[
  {
    "id": "2025_human_panel",
    "displayName": "Human Panel",
    "modelReleaseDate": null,
    "providerId": "Human",
    "modelType": null,
    "modelGroup": "Human",
    "paperUrl": null,
    "codeUrl": null,
    "dataUrl": null,
    "featured": null
  },
  {
    "id": "ARChitects",
    "displayName": "ARChitects",
    "modelReleaseDate": "2024-11-03T00:00:00.000Z",
    "providerId": "ARC Prize 2024",
    "modelType": "Custom",
    "modelGroup": "Kaggle",
    "paperUrl": "https://github.com/da-fr/arc-prize-2024/blob/main/the_architects.pdf",
    "codeUrl": "https://www.kaggle.com/code/gregkamradt/arc-prize-v8?scriptVersionId=211457842",
    "dataUrl": null,
    "featured": null
  },
  {
    "id": "NVARC_2025",
    "displayName": "NVARC",
    "modelReleaseDate": "2024-11-03T00:00:00.000Z",
    "providerId": "ARC Prize 2025",
    "modelType": "Custom",
    "modelGroup": "Kaggle",
    "paperUrl": "https://drive.google.com/file/d/1vkEluaaJTzaZiJL69TkZovJUkPSDH5Xc/view",
    "codeUrl": "https://www.kaggle.com/code/gregkamradt/arc2-qwen3-unsloth-flash-lora-batch8-queue-trm2/edit?fromFork=1",
    "dataUrl": null,
    "featured": null
  },
  {
    "id": "Claude 3.7",
    "displayName": "Claude 3.7",
    "modelReleaseDate": "2025-02-24T00:00:00.000Z",
    "providerId": "Anthropic",
    "modelType": "Base LLM",
    "modelGroup": null,
    "paperUrl": null,
    "codeUrl": "https://github.com/arcprize/model_baseline",
    "dataUrl": null,
    "featured": null
  },
  {
    "id": "Claude 3.7 Thinking 16K",
    "displayName": "Claude 3.7 (16K)",
    "modelReleaseDate": "2025-02-24T00:00:00.000Z",
    "providerId": "Anthropic",
    "modelType": "CoT",
    "modelGroup": "Claude_3_7_thinking",
    "paperUrl": null,
    "codeUrl": "https://github.com/arcprize/model_baseline",
    "dataUrl": null,
    "featured": null
  },
  {
    "id": "Claude 3.7 Thinking 1K",
    "displayName": "Claude 3.7 (1K)",
    "modelReleaseDate": "2025-02-24T00:00:00.000Z",
    "providerId": "Anthropic",
    "modelType": "CoT",
    "modelGroup": "Claude_3_7_thinking",
    "paperUrl": null,
    "codeUrl": "https://github.com/arcprize/model_baseline",
    "dataUrl": null,
    "featured": null
  },
  {
    "id": "Claude 3.7 Thinking 8K",
    "displayName": "Claude 3.7 (8K)",
    "modelReleaseDate": "2025-02-24T00:00:00.000Z",
    "providerId": "Anthropic",
    "modelType": "CoT",
    "modelGroup": "Claude_3_7_thinking",
    "paperUrl": null,
    "codeUrl": "https://github.com/arcprize/model_baseline",
    "dataUrl": null,
    "featured": null
  },
  {
    "id": "Gemini 1.5 Pro",
    "displayName": "Gemini 1.5 Pro",
    "modelReleaseDate": "2025-02-15T00:00:00.000Z",
    "providerId": "Google",
    "modelType": "Base LLM",
    "modelGroup": null,
    "paperUrl": null,
    "codeUrl": "https://github.com/arcprize/model_baseline",
    "dataUrl": null,
    "featured": null
  },
  {
    "id": "Gemini 2.0 Flash",
    "displayName": "Gemini 2.0 Flash",
    "modelReleaseDate": "2025-02-02T00:00:00.000Z",
    "providerId": "Google",
    "modelType": "Base LLM",
    "modelGroup": null,
    "paperUrl": null,
    "codeUrl": "https://github.com/arcprize/model_baseline",
    "dataUrl": null,
    "featured": null
  },
  {
    "id": "gpt-4-5-2025-02-27",
    "displayName": "GPT-4.5",
    "modelReleaseDate": "2025-02-27T00:00:00.000Z",
    "providerId": "OpenAI",
    "modelType": "Base LLM",
    "modelGroup": null,
    "paperUrl": null,
    "codeUrl": "https://github.com/arcprize/model_baseline",
    "dataUrl": null,
    "featured": null
  },
  {
    "id": "gpt-4o-2024-11-20",
    "displayName": "GPT-4o",
    "modelReleaseDate": "2024-11-20T00:00:00.000Z",
    "providerId": "OpenAI",
    "modelType": "Base LLM",
    "modelGroup": null,
    "paperUrl": null,
    "codeUrl": "https://github.com/arcprize/model_baseline",
    "dataUrl": null,
    "featured": null
  },
  {
    "id": "gpt-4o-mini-2024-07-18",
    "displayName": "GPT-4o-mini",
    "modelReleaseDate": "2024-07-18T00:00:00.000Z",
    "providerId": "OpenAI",
    "modelType": "Base LLM",
    "modelGroup": null,
    "paperUrl": null,
    "codeUrl": "https://github.com/arcprize/model_baseline",
    "dataUrl": null,
    "featured": null
  },
  {
    "id": "Grok 3",
    "displayName": "Grok 3",
    "modelReleaseDate": "2025-02-13T00:00:00.000Z",
    "providerId": "xAI",
    "modelType": "Base LLM",
    "modelGroup": null,
    "paperUrl": null,
    "codeUrl": null,
    "dataUrl": null,
    "featured": null
  },
  {
    "id": "Grok 3 Thinking",
    "displayName": "Grok 3 Thinking",
    "modelReleaseDate": "2025-02-13T00:00:00.000Z",
    "providerId": "xAI",
    "modelType": "CoT",
    "modelGroup": null,
    "paperUrl": null,
    "codeUrl": null,
    "dataUrl": null,
    "featured": null
  },
  {
    "id": "Icecuber",
    "displayName": "Icecuber",
    "modelReleaseDate": "2023-11-03T00:00:00.000Z",
    "providerId": "ARC Prize 2024",
    "modelType": "Custom",
    "modelGroup": "Kaggle",
    "paperUrl": null,
    "codeUrl": "https://www.kaggle.com/code/hansuelijud/template-arc2020-1st-place-solution-by-icecuber",
    "dataUrl": null,
    "featured": null
  },
  {
    "id": "Llama 3.3",
    "displayName": "Llama 3.3",
    "modelReleaseDate": "2024-12-06T00:00:00.000Z",
    "providerId": "Meta",
    "modelType": "Base LLM",
    "modelGroup": null,
    "paperUrl": null,
    "codeUrl": null,
    "dataUrl": null,
    "featured": null
  },
  {
    "id": "mturker",
    "displayName": "Avg. Mturker",
    "modelReleaseDate": null,
    "providerId": "Human",
    "modelType": null,
    "modelGroup": "Human",
    "paperUrl": null,
    "codeUrl": null,
    "dataUrl": null,
    "featured": null
  },
  {
    "id": "o1 - high",
    "displayName": "o1 (High)",
    "modelReleaseDate": "2024-12-05T00:00:00.000Z",
    "providerId": "OpenAI",
    "modelType": "CoT",
    "modelGroup": "o1",
    "paperUrl": null,
    "codeUrl": "https://github.com/arcprize/model_baseline",
    "dataUrl": null,
    "featured": null
  },
  {
    "id": "o1 - medium",
    "displayName": "o1 (Medium)",
    "modelReleaseDate": "2024-12-05T00:00:00.000Z",
    "providerId": "OpenAI",
    "modelType": "CoT",
    "modelGroup": "o1",
    "paperUrl": null,
    "codeUrl": "https://github.com/arcprize/model_baseline",
    "dataUrl": null,
    "featured": null
  },
  {
    "id": "o1-pro",
    "displayName": "o1-pro",
    "modelReleaseDate": "2024-12-05T00:00:00.000Z",
    "providerId": "OpenAI",
    "modelType": "CoT + Synthesis",
    "modelGroup": "o1-pro",
    "paperUrl": null,
    "codeUrl": null,
    "dataUrl": null,
    "featured": null
  },
  {
    "id": "o1-preview-2024-09-12",
    "displayName": "o1-preview",
    "modelReleaseDate": "2024-09-12T00:00:00.000Z",
    "providerId": "OpenAI",
    "modelType": "CoT",
    "modelGroup": "o1-preview",
    "paperUrl": "https://openai.com/index/introducing-openai-o1-preview/",
    "codeUrl": null,
    "dataUrl": null,
    "featured": null
  },
  {
    "id": "o3-preview-low",
    "displayName": "o3 (Preview, Low) ¹",
    "modelReleaseDate": "2024-12-20T00:00:00.000Z",
    "providerId": "OpenAI",
    "modelType": "CoT + Synthesis",
    "modelGroup": "o3-preview",
    "paperUrl": "https://arcprize.org/blog/oai-o3-pub-breakthrough",
    "codeUrl": null,
    "dataUrl": null,
    "featured": null
  },
  {
    "id": "o3-preview-high",
    "displayName": "o3 (Preview, High)*",
    "modelReleaseDate": "2024-12-20T00:00:00.000Z",
    "providerId": "OpenAI",
    "modelType": "CoT + Synthesis",
    "modelGroup": "o3-preview",
    "paperUrl": "https://arcprize.org/blog/oai-o3-pub-breakthrough",
    "codeUrl": null,
    "dataUrl": null,
    "featured": null
  },
  {
    "id": "o3-mini-high",
    "displayName": "o3-mini (High)",
    "modelReleaseDate": "2025-01-31T00:00:00.000Z",
    "providerId": "OpenAI",
    "modelType": "CoT",
    "modelGroup": "o3-mini",
    "paperUrl": null,
    "codeUrl": "https://github.com/arcprize/model_baseline",
    "dataUrl": null,
    "featured": null
  },
  {
    "id": "o3-mini-medium",
    "displayName": "o3-mini (Medium)",
    "modelReleaseDate": "2025-01-31T00:00:00.000Z",
    "providerId": "OpenAI",
    "modelType": "CoT",
    "modelGroup": "o3-mini",
    "paperUrl": null,
    "codeUrl": "https://github.com/arcprize/model_baseline",
    "dataUrl": null,
    "featured": null
  },
  {
    "id": "o3-mini-low",
    "displayName": "o3-mini (Low)",
    "modelReleaseDate": "2025-01-31T00:00:00.000Z",
    "providerId": "OpenAI",
    "modelType": "CoT",
    "modelGroup": "o3-mini",
    "paperUrl": null,
    "codeUrl": "https://github.com/arcprize/model_baseline",
    "dataUrl": null,
    "featured": null
  },
  {
    "id": "R1",
    "displayName": "Deepseek R1",
    "modelReleaseDate": "2025-01-22T00:00:00.000Z",
    "providerId": "DeepSeek",
    "modelType": "CoT",
    "modelGroup": null,
    "paperUrl": null,
    "codeUrl": "https://github.com/arcprize/model_baseline",
    "dataUrl": null,
    "featured": null
  },
  {
    "id": "stem_grad",
    "displayName": "Stem Grad",
    "modelReleaseDate": null,
    "providerId": "Human",
    "modelType": null,
    "modelGroup": "Human",
    "paperUrl": null,
    "codeUrl": null,
    "dataUrl": null,
    "featured": null
  },
  {
    "id": "gemini-2-5-flash-preview-05-20",
    "displayName": "Gemini 2.5 Flash (Preview)",
    "modelReleaseDate": "2025-05-20T00:00:00.000Z",
    "providerId": "Google",
    "modelType": "CoT",
    "modelGroup": null,
    "paperUrl": "https://deepmind.google/models/gemini/flash/",
    "codeUrl": "https://github.com/arcprize/model_baseline",
    "dataUrl": null,
    "featured": null
  },
  {
    "id": "gemini-2-5-flash-preview-05-20-thinking-1k",
    "displayName": "Gemini 2.5 Flash (Preview) (Thinking 1K)",
    "modelReleaseDate": "2025-05-20T00:00:00.000Z",
    "providerId": "Google",
    "modelType": "CoT",
    "modelGroup": "gemini-2-5-flash-preview-05-20-thinking",
    "paperUrl": "https://deepmind.google/models/gemini/flash/",
    "codeUrl": "https://github.com/arcprize/model_baseline",
    "dataUrl": null,
    "featured": null
  },
  {
    "id": "gemini-2-5-flash-preview-05-20-thinking-8k",
    "displayName": "Gemini 2.5 Flash (Preview) (Thinking 8K)",
    "modelReleaseDate": "2025-05-20T00:00:00.000Z",
    "providerId": "Google",
    "modelType": "CoT",
    "modelGroup": "gemini-2-5-flash-preview-05-20-thinking",
    "paperUrl": "https://deepmind.google/models/gemini/flash/",
    "codeUrl": "https://github.com/arcprize/model_baseline",
    "dataUrl": null,
    "featured": null
  },
  {
    "id": "gemini-2-5-flash-preview-05-20-thinking-16k",
    "displayName": "Gemini 2.5 Flash (Preview) (Thinking 16K)",
    "modelReleaseDate": "2025-05-20T00:00:00.000Z",
    "providerId": "Google",
    "modelType": "CoT",
    "modelGroup": "gemini-2-5-flash-preview-05-20-thinking",
    "paperUrl": "https://deepmind.google/models/gemini/flash/",
    "codeUrl": "https://github.com/arcprize/model_baseline",
    "dataUrl": null,
    "featured": null
  },
  {
    "id": "gemini-2-5-flash-preview-05-20-thinking-24k",
    "displayName": "Gemini 2.5 Flash (Preview) (Thinking 24K)",
    "modelReleaseDate": "2025-05-20T00:00:00.000Z",
    "providerId": "Google",
    "modelType": "CoT",
    "modelGroup": "gemini-2-5-flash-preview-05-20-thinking",
    "paperUrl": "https://deepmind.google/models/gemini/flash/",
    "codeUrl": "https://github.com/arcprize/model_baseline",
    "dataUrl": null,
    "featured": null
  },
  {
    "id": "gemini-2.5-pro-exp-03-25",
    "displayName": "Gemini-2.5-Pro-Exp-03-25 **",
    "modelReleaseDate": "2025-03-25T00:00:00.000Z",
    "providerId": "Google",
    "modelType": "CoT",
    "modelGroup": null,
    "paperUrl": null,
    "codeUrl": "https://github.com/arcprize/model_baseline",
    "dataUrl": null,
    "featured": null
  },
  {
    "id": "Llama-4-Maverick-17B-128E-Instruct-FP8-together",
    "displayName": "Llama 4 Maverick",
    "modelReleaseDate": "2025-04-05T00:00:00.000Z",
    "providerId": "Meta",
    "modelType": "Base LLM",
    "modelGroup": "Llama 4",
    "paperUrl": "https://www.llama.com/docs/model-cards-and-prompt-formats/llama4_omni/",
    "codeUrl": "https://github.com/arcprize/model_baseline",
    "dataUrl": null,
    "featured": null
  },
  {
    "id": "Llama-4-Scout-17B-16E-Instruct-together",
    "displayName": "Llama 4 Scout",
    "modelReleaseDate": "2025-04-05T00:00:00.000Z",
    "providerId": "Meta",
    "modelType": "Base LLM",
    "modelGroup": "Llama 4",
    "paperUrl": "https://www.llama.com/docs/model-cards-and-prompt-formats/llama4_omni/",
    "codeUrl": "https://github.com/arcprize/model_baseline",
    "dataUrl": null,
    "featured": null
  },
  {
    "id": "gpt-4-1-nano-2025-04-14",
    "displayName": "GPT-4.1-Nano",
    "modelReleaseDate": "2025-04-14T00:00:00.000Z",
    "providerId": "OpenAI",
    "modelType": "Base LLM",
    "modelGroup": "gpt-4.1",
    "paperUrl": "https://openai.com/index/gpt-4-1/",
    "codeUrl": "https://github.com/arcprize/model_baseline",
    "dataUrl": null,
    "featured": null
  },
  {
    "id": "gpt-4-1-mini-2025-04-14",
    "displayName": "GPT-4.1-Mini",
    "modelReleaseDate": "2025-04-14T00:00:00.000Z",
    "providerId": "OpenAI",
    "modelType": "Base LLM",
    "modelGroup": "gpt-4.1",
    "paperUrl": "https://openai.com/index/gpt-4-1/",
    "codeUrl": "https://github.com/arcprize/model_baseline",
    "dataUrl": null,
    "featured": null
  },
  {
    "id": "gpt-4-1-2025-04-14",
    "displayName": "GPT-4.1",
    "modelReleaseDate": "2025-04-14T00:00:00.000Z",
    "providerId": "OpenAI",
    "modelType": "Base LLM",
    "modelGroup": "gpt-4.1",
    "paperUrl": "https://openai.com/index/gpt-4-1/",
    "codeUrl": "https://github.com/arcprize/model_baseline",
    "dataUrl": null,
    "featured": null
  },
  {
    "id": "o1-mini-2024-09-12",
    "displayName": "o1-mini",
    "modelReleaseDate": "2024-09-12T00:00:00.000Z",
    "providerId": "OpenAI",
    "modelType": "CoT",
    "modelGroup": "o1-mini",
    "paperUrl": "https://openai.com/index/openai-o1-mini-advancing-cost-efficient-reasoning/",
    "codeUrl": "https://github.com/arcprize/model_baseline",
    "dataUrl": null,
    "featured": null
  },
  {
    "id": "o1-pro-2025-03-19-low",
    "displayName": "o1-pro (Low)",
    "modelReleaseDate": "2025-03-19T00:00:00.000Z",
    "providerId": "OpenAI",
    "modelType": "CoT + Synthesis",
    "modelGroup": "o1-pro",
    "paperUrl": "https://platform.openai.com/docs/models/o1-pro",
    "codeUrl": "https://github.com/arcprize/model_baseline",
    "dataUrl": null,
    "featured": null
  },
  {
    "id": "o1-2024-12-17-low",
    "displayName": "o1 (Low)",
    "modelReleaseDate": "2024-12-17T00:00:00.000Z",
    "providerId": "OpenAI",
    "modelType": "CoT",
    "modelGroup": "o1",
    "paperUrl": "https://openai.com/o1/",
    "codeUrl": "https://github.com/arcprize/model_baseline",
    "dataUrl": null,
    "featured": null
  },
  {
    "id": "o1-2024-12-17-medium",
    "displayName": "o1 (Medium)",
    "modelReleaseDate": "2024-12-17T00:00:00.000Z",
    "providerId": "OpenAI",
    "modelType": "CoT",
    "modelGroup": "o1",
    "paperUrl": "https://openai.com/o1/",
    "codeUrl": "https://github.com/arcprize/model_baseline",
    "dataUrl": null,
    "featured": null
  },
  {
    "id": "o1-2024-12-17-high",
    "displayName": "o1 (High)",
    "modelReleaseDate": "2024-12-17T00:00:00.000Z",
    "providerId": "OpenAI",
    "modelType": "CoT",
    "modelGroup": "o1",
    "paperUrl": "https://openai.com/o1/",
    "codeUrl": "https://github.com/arcprize/model_baseline",
    "dataUrl": null,
    "featured": null
  },
  {
    "id": "o3-mini-2025-01-31-low",
    "displayName": "o3-mini (Low)",
    "modelReleaseDate": "2025-01-31T00:00:00.000Z",
    "providerId": "OpenAI",
    "modelType": "CoT",
    "modelGroup": "o3-mini",
    "paperUrl": "",
    "codeUrl": "",
    "dataUrl": null,
    "featured": null
  },
  {
    "id": "o3-2025-04-16-high",
    "displayName": "o3 (High)",
    "modelReleaseDate": "2025-04-16T00:00:00.000Z",
    "providerId": "OpenAI",
    "modelType": "CoT",
    "modelGroup": "o3",
    "paperUrl": "",
    "codeUrl": "",
    "dataUrl": null,
    "featured": null
  },
  {
    "id": "o3-2025-04-16-low",
    "displayName": "o3 (Low)",
    "modelReleaseDate": "2025-04-16T00:00:00.000Z",
    "providerId": "OpenAI",
    "modelType": "CoT",
    "modelGroup": "o3",
    "paperUrl": "",
    "codeUrl": "",
    "dataUrl": null,
    "featured": null
  },
  {
    "id": "o3-2025-04-16-medium",
    "displayName": "o3 (Medium)",
    "modelReleaseDate": "2025-04-16T00:00:00.000Z",
    "providerId": "OpenAI",
    "modelType": "CoT",
    "modelGroup": "o3",
    "paperUrl": "",
    "codeUrl": "",
    "dataUrl": null,
    "featured": null
  },
  {
    "id": "o3-mini-2025-01-31-high",
    "displayName": "o3-mini (High)",
    "modelReleaseDate": "2025-01-31T00:00:00.000Z",
    "providerId": "OpenAI",
    "modelType": "CoT",
    "modelGroup": "o3-mini",
    "paperUrl": "",
    "codeUrl": "",
    "dataUrl": null,
    "featured": null
  },
  {
    "id": "o3-mini-2025-01-31-medium",
    "displayName": "o3-mini (Medium)",
    "modelReleaseDate": "2025-01-31T00:00:00.000Z",
    "providerId": "OpenAI",
    "modelType": "CoT",
    "modelGroup": "o3-mini",
    "paperUrl": "",
    "codeUrl": "",
    "dataUrl": null,
    "featured": null
  },
  {
    "id": "o4-mini-2025-04-16-high",
    "displayName": "o4-mini (High)",
    "modelReleaseDate": "2025-04-16T00:00:00.000Z",
    "providerId": "OpenAI",
    "modelType": "CoT",
    "modelGroup": "o4-mini",
    "paperUrl": "",
    "codeUrl": "",
    "dataUrl": null,
    "featured": null
  },
  {
    "id": "o4-mini-2025-04-16-low",
    "displayName": "o4-mini (Low)",
    "modelReleaseDate": "2025-04-16T00:00:00.000Z",
    "providerId": "OpenAI",
    "modelType": "CoT",
    "modelGroup": "o4-mini",
    "paperUrl": "",
    "codeUrl": "",
    "dataUrl": null,
    "featured": null
  },
  {
    "id": "o4-mini-2025-04-16-medium",
    "displayName": "o4-mini (Medium)",
    "modelReleaseDate": "2025-04-16T00:00:00.000Z",
    "providerId": "OpenAI",
    "modelType": "CoT",
    "modelGroup": "o4-mini",
    "paperUrl": "",
    "codeUrl": "",
    "dataUrl": null,
    "featured": null
  },
  {
    "id": "codex-mini-latest",
    "displayName": "Codex Mini (Latest)",
    "modelReleaseDate": "2025-05-23T00:00:00.000Z",
    "providerId": "OpenAI",
    "modelType": "CoT",
    "modelGroup": "",
    "paperUrl": "https://platform.openai.com/docs/models/codex-mini-latest",
    "codeUrl": "",
    "dataUrl": null,
    "featured": null
  },
  {
    "id": "claude-sonnet-4-20250514",
    "displayName": "Claude Sonnet 4",
    "modelReleaseDate": "2025-05-14T00:00:00.000Z",
    "providerId": "Anthropic",
    "modelType": "Base LLM",
    "modelGroup": "",
    "paperUrl": "https://www.anthropic.com/news/claude-4",
    "codeUrl": "",
    "dataUrl": null,
    "featured": null
  },
  {
    "id": "claude-sonnet-4-20250514-thinking-1k",
    "displayName": "Claude Sonnet 4 (Thinking 1K)",
    "modelReleaseDate": "2025-05-14T00:00:00.000Z",
    "providerId": "Anthropic",
    "modelType": "CoT",
    "modelGroup": "claude-sonnet-4-20250514-thinking",
    "paperUrl": "https://www.anthropic.com/news/claude-4",
    "codeUrl": "",
    "dataUrl": null,
    "featured": null
  },
  {
    "id": "claude-sonnet-4-20250514-thinking-8k-bedrock",
    "displayName": "Claude Sonnet 4 (Thinking 8K)",
    "modelReleaseDate": "2025-05-14T00:00:00.000Z",
    "providerId": "Anthropic",
    "modelType": "CoT",
    "modelGroup": "claude-sonnet-4-20250514-thinking",
    "paperUrl": "https://www.anthropic.com/news/claude-4",
    "codeUrl": "",
    "dataUrl": null,
    "featured": null
  },
  {
    "id": "claude-sonnet-4-20250514-thinking-16k-bedrock",
    "displayName": "Claude Sonnet 4 (Thinking 16K)",
    "modelReleaseDate": "2025-05-14T00:00:00.000Z",
    "providerId": "Anthropic",
    "modelType": "CoT",
    "modelGroup": "claude-sonnet-4-20250514-thinking",
    "paperUrl": "https://www.anthropic.com/news/claude-4",
    "codeUrl": "",
    "dataUrl": null,
    "featured": null
  },
  {
    "id": "claude-opus-4-20250514",
    "displayName": "Claude Opus 4",
    "modelReleaseDate": "2025-05-14T00:00:00.000Z",
    "providerId": "Anthropic",
    "modelType": "CoT",
    "modelGroup": "",
    "paperUrl": "https://www.anthropic.com/news/claude-4",
    "codeUrl": "",
    "dataUrl": null,
    "featured": null
  },
  {
    "id": "claude-opus-4-20250514-thinking-1k",
    "displayName": "Claude Opus 4 (Thinking 1K)",
    "modelReleaseDate": "2025-05-14T00:00:00.000Z",
    "providerId": "Anthropic",
    "modelType": "CoT",
    "modelGroup": "claude-opus-4-20250514-thinking",
    "paperUrl": "https://www.anthropic.com/news/claude-4",
    "codeUrl": "",
    "dataUrl": null,
    "featured": null
  },
  {
    "id": "claude-opus-4-20250514-thinking-8k",
    "displayName": "Claude Opus 4 (Thinking 8K)",
    "modelReleaseDate": "2025-05-14T00:00:00.000Z",
    "providerId": "Anthropic",
    "modelType": "CoT",
    "modelGroup": "claude-opus-4-20250514-thinking",
    "paperUrl": "https://www.anthropic.com/news/claude-4",
    "codeUrl": "",
    "dataUrl": null,
    "featured": null
  },
  {
    "id": "claude-opus-4-20250514-thinking-16k",
    "displayName": "Claude Opus 4 (Thinking 16K)",
    "modelReleaseDate": "2025-05-14T00:00:00.000Z",
    "providerId": "Anthropic",
    "modelType": "CoT",
    "modelGroup": "claude-opus-4-20250514-thinking",
    "paperUrl": "https://www.anthropic.com/news/claude-4",
    "codeUrl": "",
    "dataUrl": null,
    "featured": null
  },
  {
    "id": "deepseek_r1_0528-openrouter",
    "displayName": "Deepseek R1 (05/28)",
    "modelReleaseDate": "2025-05-28T00:00:00.000Z",
    "providerId": "DeepSeek",
    "modelType": "CoT",
    "modelGroup": "",
    "paperUrl": "https://api-docs.deepseek.com/news/news250528",
    "codeUrl": "",
    "dataUrl": null,
    "featured": null
  },
  {
    "id": "grok-4-0709",
    "displayName": "Grok 4 (Thinking)",
    "modelReleaseDate": "2025-07-09T00:00:00.000Z",
    "providerId": "xAI",
    "modelType": "CoT",
    "modelGroup": "",
    "paperUrl": "https://x.ai/news",
    "codeUrl": "",
    "dataUrl": null,
    "featured": null
  },
  {
    "id": "grok-4-fast-reasoning",
    "displayName": "Grok 4 (Fast Reasoning)",
    "modelReleaseDate": "2025-10-21T00:00:00.000Z",
    "providerId": "xAI",
    "modelType": "CoT",
    "modelGroup": "grok-4-0709",
    "paperUrl": "https://x.ai/news/grok-4-fast",
    "codeUrl": "",
    "dataUrl": "",
    "featured": null
  },
  {
    "id": "grok-3-openrouter",
    "displayName": "Grok 3",
    "modelReleaseDate": "2025-02-19T00:00:00.000Z",
    "providerId": "xAI",
    "modelType": "Base LLM",
    "modelGroup": "",
    "paperUrl": "https://x.ai/api#capabilities",
    "codeUrl": "",
    "dataUrl": null,
    "featured": null
  },
  {
    "id": "grok-3-mini-beta-low-openrouter",
    "displayName": "Grok 3 Mini (Low)",
    "modelReleaseDate": "2025-02-19T00:00:00.000Z",
    "providerId": "xAI",
    "modelType": "CoT",
    "modelGroup": "",
    "paperUrl": "https://x.ai/api#capabilities",
    "codeUrl": "",
    "dataUrl": null,
    "featured": null
  },
  {
    "id": "gemini-2-5-pro-preview-openrouter",
    "displayName": "Gemini 2.5 Pro (Preview)",
    "modelReleaseDate": "2025-02-19T00:00:00.000Z",
    "providerId": "Google",
    "modelType": "CoT",
    "modelGroup": "",
    "paperUrl": "https://blog.google/technology/google-deepmind/gemini-model-thinking-updates-march-2025/",
    "codeUrl": "",
    "dataUrl": null,
    "featured": null
  },
  {
    "id": "gemini-2-5-pro-preview-05-06-thinking-1k",
    "displayName": "Gemini 2.5 Pro (Preview, Thinking 1K)",
    "modelReleaseDate": "2025-02-19T00:00:00.000Z",
    "providerId": "Google",
    "modelType": "CoT",
    "modelGroup": "",
    "paperUrl": "https://blog.google/technology/google-deepmind/gemini-model-thinking-updates-march-2025/",
    "codeUrl": "",
    "dataUrl": null,
    "featured": null
  },
  {
    "id": "o3-pro-2025-06-10-high",
    "displayName": "o3-Pro (High)",
    "modelReleaseDate": "2025-06-10T00:00:00.000Z",
    "providerId": "OpenAI",
    "modelType": "CoT + Synthesis",
    "modelGroup": "o3-pro",
    "paperUrl": "https://blog.google/technology/google-deepmind/gemini-model-thinking-updates-march-2025/",
    "codeUrl": "",
    "dataUrl": null,
    "featured": null
  },
  {
    "id": "o3-pro-2025-06-10-medium",
    "displayName": "o3-Pro (Medium)",
    "modelReleaseDate": "2025-06-10T00:00:00.000Z",
    "providerId": "OpenAI",
    "modelType": "CoT + Synthesis",
    "modelGroup": "o3-pro",
    "paperUrl": "https://blog.google/technology/google-deepmind/gemini-model-thinking-updates-march-2025/",
    "codeUrl": "",
    "dataUrl": null,
    "featured": null
  },
  {
    "id": "o3-pro-2025-06-10-low",
    "displayName": "o3-Pro (Low)",
    "modelReleaseDate": "2025-06-10T00:00:00.000Z",
    "providerId": "OpenAI",
    "modelType": "CoT + Synthesis",
    "modelGroup": "o3-pro",
    "paperUrl": "https://blog.google/technology/google-deepmind/gemini-model-thinking-updates-march-2025/",
    "codeUrl": "",
    "dataUrl": null,
    "featured": null
  },
  {
    "id": "magistral-small-2506",
    "displayName": "Magistral Small",
    "modelReleaseDate": "2025-06-10T00:00:00.000Z",
    "providerId": "Mistral",
    "modelType": "CoT",
    "modelGroup": "",
    "paperUrl": "https://mistral.ai/news/magistral",
    "codeUrl": "",
    "dataUrl": null,
    "featured": null
  },
  {
    "id": "magistral-medium-2506",
    "displayName": "Magistral Medium",
    "modelReleaseDate": "2025-06-10T00:00:00.000Z",
    "providerId": "Mistral",
    "modelType": "CoT",
    "modelGroup": "",
    "paperUrl": "https://mistral.ai/news/magistral",
    "codeUrl": "",
    "dataUrl": null,
    "featured": null
  },
  {
    "id": "magistral-medium-2506-thinking",
    "displayName": "Magistral Medium (Thinking)",
    "modelReleaseDate": "2025-06-10T00:00:00.000Z",
    "providerId": "Mistral",
    "modelType": "CoT",
    "modelGroup": "",
    "paperUrl": "https://mistral.ai/news/magistral",
    "codeUrl": "",
    "dataUrl": null,
    "featured": null
  },
  {
    "id": "gemini-2-5-pro-2025-06-17-thinking-1k",
    "displayName": "Gemini 2.5 Pro (Thinking 1K)",
    "modelReleaseDate": "2025-06-17T00:00:00.000Z",
    "providerId": "Google",
    "modelType": "CoT",
    "modelGroup": "gemini-2-5-pro-2025-06-17-thinking",
    "paperUrl": "https://cloud.google.com/blog/products/ai-machine-learning/gemini-2-5-flash-lite-flash-pro-ga-vertex-ai",
    "codeUrl": "",
    "dataUrl": null,
    "featured": null
  },
  {
    "id": "gemini-2-5-pro-2025-06-17-thinking-8k",
    "displayName": "Gemini 2.5 Pro (Thinking 8K)",
    "modelReleaseDate": "2025-06-17T00:00:00.000Z",
    "providerId": "Google",
    "modelType": "CoT",
    "modelGroup": "gemini-2-5-pro-2025-06-17-thinking",
    "paperUrl": "https://cloud.google.com/blog/products/ai-machine-learning/gemini-2-5-flash-lite-flash-pro-ga-vertex-ai",
    "codeUrl": "",
    "dataUrl": null,
    "featured": null
  },
  {
    "id": "gemini-2-5-pro-2025-06-17-thinking-16k",
    "displayName": "Gemini 2.5 Pro (Thinking 16K)",
    "modelReleaseDate": "2025-06-17T00:00:00.000Z",
    "providerId": "Google",
    "modelType": "CoT",
    "modelGroup": "gemini-2-5-pro-2025-06-17-thinking",
    "paperUrl": "https://cloud.google.com/blog/products/ai-machine-learning/gemini-2-5-flash-lite-flash-pro-ga-vertex-ai",
    "codeUrl": "",
    "dataUrl": null,
    "featured": null
  },
  {
    "id": "gemini-2-5-pro-2025-06-17-thinking-32k",
    "displayName": "Gemini 2.5 Pro (Thinking 32K)",
    "modelReleaseDate": "2025-06-17T00:00:00.000Z",
    "providerId": "Google",
    "modelType": "CoT",
    "modelGroup": "gemini-2-5-pro-2025-06-17-thinking",
    "paperUrl": "https://cloud.google.com/blog/products/ai-machine-learning/gemini-2-5-flash-lite-flash-pro-ga-vertex-ai",
    "codeUrl": "",
    "dataUrl": null,
    "featured": null
  },
  {
    "id": "qwen3-235b-a22b-instruct-2507",
    "displayName": "Qwen3-235b-a22b Instruct (25/07)",
    "modelReleaseDate": "2025-07-25T00:00:00.000Z",
    "providerId": "Alibaba",
    "modelType": "Base LLM",
    "modelGroup": "",
    "paperUrl": "https://huggingface.co/Qwen/Qwen3-235B-A22B-Instruct-2507",
    "codeUrl": "",
    "dataUrl": "https://huggingface.co/datasets/arcprize/arc_agi_v2_public_eval/tree/main/qwen3-235b-a22b-instruct-2507",
    "featured": null
  },
  {
    "id": "gpt-5-2025-08-07-minimal",
    "displayName": "GPT-5 (Minimal)",
    "modelReleaseDate": "2025-08-07T00:00:00.000Z",
    "providerId": "OpenAI",
    "modelType": "Base LLM",
    "modelGroup": "",
    "paperUrl": "",
    "codeUrl": "",
    "dataUrl": "",
    "featured": null
  },
  {
    "id": "gpt-5-2025-08-07-low",
    "displayName": "GPT-5 (Low)",
    "modelReleaseDate": "2025-08-07T00:00:00.000Z",
    "providerId": "OpenAI",
    "modelType": "CoT",
    "modelGroup": "gpt-5-reasoning",
    "paperUrl": "",
    "codeUrl": "",
    "dataUrl": "",
    "featured": null
  },
  {
    "id": "gpt-5-2025-08-07-medium",
    "displayName": "GPT-5 (Medium)",
    "modelReleaseDate": "2025-08-07T00:00:00.000Z",
    "providerId": "OpenAI",
    "modelType": "CoT",
    "modelGroup": "gpt-5-reasoning",
    "paperUrl": "",
    "codeUrl": "",
    "dataUrl": "",
    "featured": null
  },
  {
    "id": "gpt-5-2025-08-07-high",
    "displayName": "GPT-5 (High)",
    "modelReleaseDate": "2025-08-07T00:00:00.000Z",
    "providerId": "OpenAI",
    "modelType": "CoT",
    "modelGroup": "gpt-5-reasoning",
    "paperUrl": "",
    "codeUrl": "",
    "dataUrl": "",
    "featured": null
  },
  {
    "id": "gpt-5-1-2025-11-13-thinking-none",
    "displayName": "GPT-5.1 (Thinking, None)",
    "modelReleaseDate": "2025-11-13T00:00:00.000Z",
    "providerId": "OpenAI",
    "modelType": "CoT",
    "modelGroup": "gpt-5-1-reasoning",
    "paperUrl": "",
    "codeUrl": "",
    "dataUrl": "",
    "featured": null
  },
  {
    "id": "gpt-5-1-2025-11-13-thinking-low",
    "displayName": "GPT-5.1 (Thinking, Low)",
    "modelReleaseDate": "2025-11-13T00:00:00.000Z",
    "providerId": "OpenAI",
    "modelType": "CoT",
    "modelGroup": "gpt-5-1-reasoning",
    "paperUrl": "",
    "codeUrl": "",
    "dataUrl": "",
    "featured": null
  },
  {
    "id": "gpt-5-1-2025-11-13-thinking-medium",
    "displayName": "GPT-5.1 (Thinking, Medium)",
    "modelReleaseDate": "2025-11-13T00:00:00.000Z",
    "providerId": "OpenAI",
    "modelType": "CoT",
    "modelGroup": "gpt-5-1-reasoning",
    "paperUrl": "",
    "codeUrl": "",
    "dataUrl": "",
    "featured": null
  },
  {
    "id": "gpt-5-1-2025-11-13-thinking-high",
    "displayName": "GPT-5.1 (Thinking, High)",
    "modelReleaseDate": "2025-11-13T00:00:00.000Z",
    "providerId": "OpenAI",
    "modelType": "CoT",
    "modelGroup": "gpt-5-1-reasoning",
    "paperUrl": "",
    "codeUrl": "",
    "dataUrl": "",
    "featured": null
  },
  {
    "id": "gpt-5-mini-2025-08-07-high",
    "displayName": "GPT-5 Mini (High)",
    "modelReleaseDate": "2025-08-07T00:00:00.000Z",
    "providerId": "OpenAI",
    "modelType": "CoT",
    "modelGroup": "gpt-5-mini-reasoning",
    "paperUrl": "",
    "codeUrl": "",
    "dataUrl": "",
    "featured": null
  },
  {
    "id": "gpt-5-mini-2025-08-07-medium",
    "displayName": "GPT-5 Mini (Medium)",
    "modelReleaseDate": "2025-08-07T00:00:00.000Z",
    "providerId": "OpenAI",
    "modelType": "CoT",
    "modelGroup": "gpt-5-mini-reasoning",
    "paperUrl": "",
    "codeUrl": "",
    "dataUrl": "",
    "featured": null
  },
  {
    "id": "gpt-5-mini-2025-08-07-low",
    "displayName": "GPT-5 Mini (Low)",
    "modelReleaseDate": "2025-08-07T00:00:00.000Z",
    "providerId": "OpenAI",
    "modelType": "CoT",
    "modelGroup": "gpt-5-mini-reasoning",
    "paperUrl": "",
    "codeUrl": "",
    "dataUrl": "",
    "featured": null
  },
  {
    "id": "gpt-5-mini-2025-08-07-minimal",
    "displayName": "GPT-5 Mini (Minimal)",
    "modelReleaseDate": "2025-08-07T00:00:00.000Z",
    "providerId": "OpenAI",
    "modelType": "CoT",
    "modelGroup": "",
    "paperUrl": "",
    "codeUrl": "",
    "dataUrl": "",
    "featured": null
  },
  {
    "id": "gpt-5-nano-2025-08-07-high",
    "displayName": "GPT-5 Nano (High)",
    "modelReleaseDate": "2025-08-07T00:00:00.000Z",
    "providerId": "OpenAI",
    "modelType": "CoT",
    "modelGroup": "gpt-5-nano-reasoning",
    "paperUrl": "",
    "codeUrl": "",
    "dataUrl": "",
    "featured": null
  },
  {
    "id": "gpt-5-nano-2025-08-07-medium",
    "displayName": "GPT-5 Nano (Medium)",
    "modelReleaseDate": "2025-08-07T00:00:00.000Z",
    "providerId": "OpenAI",
    "modelType": "CoT",
    "modelGroup": "gpt-5-nano-reasoning",
    "paperUrl": "",
    "codeUrl": "",
    "dataUrl": "",
    "featured": null
  },
  {
    "id": "gpt-5-nano-2025-08-07-low",
    "displayName": "GPT-5 Nano (Low)",
    "modelReleaseDate": "2025-08-07T00:00:00.000Z",
    "providerId": "OpenAI",
    "modelType": "CoT",
    "modelGroup": "gpt-5-nano-reasoning",
    "paperUrl": "",
    "codeUrl": "",
    "dataUrl": "",
    "featured": null
  },
  {
    "id": "gpt-5-nano-2025-08-07-minimal",
    "displayName": "GPT-5 Nano (Minimal)",
    "modelReleaseDate": "2025-08-07T00:00:00.000Z",
    "providerId": "OpenAI",
    "modelType": "CoT",
    "modelGroup": "",
    "paperUrl": "",
    "codeUrl": "",
    "dataUrl": "",
    "featured": null
  },
  {
    "id": "hrm-2-2508",
    "displayName": "Hierarchical Reasoning Model (HRM)",
    "modelReleaseDate": "2025-08-07T00:00:00.000Z",
    "providerId": "bespoke",
    "modelType": "",
    "modelGroup": "",
    "paperUrl": "",
    "codeUrl": "",
    "dataUrl": "",
    "featured": null
  },
  {
    "id": "poetiq-arc-agi-gemini-3-d_v1",
    "displayName": "Gemini 3 Pro (Refine.)",
    "modelReleaseDate": "2025-12-04T00:00:00.000Z",
    "providerId": "poetiq",
    "modelType": "Refinement",
    "modelGroup": "",
    "paperUrl": "",
    "codeUrl": "",
    "dataUrl": "",
    "featured": null
  },
  {
    "id": "jeremy_sept_2025",
    "displayName": "Grok 4 (Refine.)",
    "modelReleaseDate": "2025-08-07T00:00:00.000Z",
    "providerId": "j_berman",
    "modelType": "Refinement",
    "modelGroup": "",
    "paperUrl": "",
    "codeUrl": "https://github.com/jerber/arc-lang-public",
    "dataUrl": "",
    "featured": null
  },
  {
    "id": "eric_pang_sept_2025",
    "displayName": "Grok 4 (Refine.)",
    "modelReleaseDate": "2025-08-07T00:00:00.000Z",
    "providerId": "e_pang",
    "modelType": "Refinement",
    "modelGroup": "",
    "paperUrl": "",
    "codeUrl": "https://github.com/epang080516/arc_agi",
    "dataUrl": "",
    "featured": null
  },
  {
    "id": "claude-sonnet-4-5-20250929",
    "displayName": "Claude Sonnet 4.5",
    "modelReleaseDate": "2025-09-29T00:00:00.000Z",
    "providerId": "Anthropic",
    "modelType": "Base LLM",
    "modelGroup": "claude-sonnet-4-5-20250929",
    "paperUrl": "https://www.anthropic.com/claude/sonnet",
    "codeUrl": "",
    "dataUrl": "https://huggingface.co/datasets/arcprize/arc_agi_v1_public_eval/tree/main/claude-sonnet-4-5-20250929",
    "featured": null
  },
  {
    "id": "claude-sonnet-4-5-20250929-thinking-1k",
    "displayName": "Claude Sonnet 4.5 (Thinking 1K)",
    "modelReleaseDate": "2025-09-29T00:00:00.000Z",
    "providerId": "Anthropic",
    "modelType": "CoT",
    "modelGroup": "claude-sonnet-4-5-20250929",
    "paperUrl": "https://www.anthropic.com/claude/sonnet",
    "codeUrl": "",
    "dataUrl": "https://huggingface.co/datasets/arcprize/arc_agi_v1_public_eval/tree/main/claude-sonnet-4-5-20250929",
    "featured": null
  },
  {
    "id": "claude-sonnet-4-5-20250929-thinking-8k",
    "displayName": "Claude Sonnet 4.5 (Thinking 8K)",
    "modelReleaseDate": "2025-09-29T00:00:00.000Z",
    "providerId": "Anthropic",
    "modelType": "CoT",
    "modelGroup": "claude-sonnet-4-5-20250929",
    "paperUrl": "https://www.anthropic.com/claude/sonnet",
    "codeUrl": "",
    "dataUrl": "https://huggingface.co/datasets/arcprize/arc_agi_v1_public_eval/tree/main/claude-sonnet-4-5-20250929",
    "featured": null
  },
  {
    "id": "claude-sonnet-4-5-20250929-thinking-16k",
    "displayName": "Claude Sonnet 4.5 (Thinking 16K)",
    "modelReleaseDate": "2025-09-29T00:00:00.000Z",
    "providerId": "Anthropic",
    "modelType": "CoT",
    "modelGroup": "claude-sonnet-4-5-20250929",
    "paperUrl": "https://www.anthropic.com/claude/sonnet",
    "codeUrl": "",
    "dataUrl": "https://huggingface.co/datasets/arcprize/arc_agi_v1_public_eval/tree/main/claude-sonnet-4-5-20250929",
    "featured": null
  },
  {
    "id": "claude-sonnet-4-5-20250929-thinking-32k",
    "displayName": "Claude Sonnet 4.5 (Thinking 32K)",
    "modelReleaseDate": "2025-09-29T00:00:00.000Z",
    "providerId": "Anthropic",
    "modelType": "CoT",
    "modelGroup": "claude-sonnet-4-5-20250929",
    "paperUrl": "https://www.anthropic.com/claude/sonnet",
    "codeUrl": "",
    "dataUrl": "https://huggingface.co/datasets/arcprize/arc_agi_v1_public_eval/tree/main/claude-sonnet-4-5-20250929",
    "featured": null
  },
  {
    "id": "gpt-5-pro-2025-10-06",
    "displayName": "GPT-5 Pro",
    "modelReleaseDate": "2025-10-06T00:00:00.000Z",
    "providerId": "OpenAI",
    "modelType": "CoT",
    "modelGroup": "",
    "paperUrl": "https://platform.openai.com/docs/models/gpt-5-pro",
    "codeUrl": "",
    "dataUrl": "",
    "featured": null
  },
  {
    "id": "trm-2025-10-07",
    "displayName": "Tiny Recursion Model (TRM)",
    "modelReleaseDate": "2025-10-07T00:00:00.000Z",
    "providerId": "bespoke",
    "modelType": "",
    "modelGroup": "",
    "paperUrl": "https://alexiajm.github.io/2025/09/29/tiny_recursive_models.html",
    "codeUrl": "https://github.com/SamsungSAILMontreal/TinyRecursiveModels",
    "dataUrl": "https://huggingface.co/arcprize/trm_arc_prize_verification",
    "featured": null
  },
  {
    "id": "claude-haiku-4-5-20251001",
    "displayName": "Claude Haiku 4.5",
    "modelReleaseDate": "2025-10-01T00:00:00.000Z",
    "providerId": "Anthropic",
    "modelType": "Base LLM",
    "modelGroup": "claude-haiku-4-5-20251001",
    "paperUrl": "https://www.anthropic.com/news/claude-haiku-4-5",
    "codeUrl": "",
    "dataUrl": "https://huggingface.co/datasets/arcprize/arc_agi_v1_public_eval/tree/main/claude-haiku-4-5-20251001",
    "featured": null
  },
  {
    "id": "claude-haiku-4-5-20251001-thinking-1k",
    "displayName": "Claude Haiku 4.5 (Thinking 1K)",
    "modelReleaseDate": "2025-10-01T00:00:00.000Z",
    "providerId": "Anthropic",
    "modelType": "CoT",
    "modelGroup": "claude-haiku-4-5-20251001",
    "paperUrl": "https://www.anthropic.com/news/claude-haiku-4-5",
    "codeUrl": "",
    "dataUrl": "https://huggingface.co/datasets/arcprize/arc_agi_v1_public_eval/tree/main/claude-haiku-4-5-20251001-thinking-1k",
    "featured": null
  },
  {
    "id": "claude-haiku-4-5-20251001-thinking-8k",
    "displayName": "Claude Haiku 4.5 (Thinking 8K)",
    "modelReleaseDate": "2025-10-01T00:00:00.000Z",
    "providerId": "Anthropic",
    "modelType": "CoT",
    "modelGroup": "claude-haiku-4-5-20251001",
    "paperUrl": "https://www.anthropic.com/news/claude-haiku-4-5",
    "codeUrl": "",
    "dataUrl": "https://huggingface.co/datasets/arcprize/arc_agi_v1_public_eval/tree/main/claude-haiku-4-5-20251001-thinking-8k",
    "featured": null
  },
  {
    "id": "claude-haiku-4-5-20251001-thinking-16k",
    "displayName": "Claude Haiku 4.5 (Thinking 16K)",
    "modelReleaseDate": "2025-10-01T00:00:00.000Z",
    "providerId": "Anthropic",
    "modelType": "CoT",
    "modelGroup": "claude-haiku-4-5-20251001",
    "paperUrl": "https://www.anthropic.com/news/claude-haiku-4-5",
    "codeUrl": "",
    "dataUrl": "https://huggingface.co/datasets/arcprize/arc_agi_v1_public_eval/tree/main/claude-haiku-4-5-20251001-thinking-16k",
    "featured": null
  },
  {
    "id": "claude-haiku-4-5-20251001-thinking-32k",
    "displayName": "Claude Haiku 4.5 (Thinking 32K)",
    "modelReleaseDate": "2025-10-01T00:00:00.000Z",
    "providerId": "Anthropic",
    "modelType": "CoT",
    "modelGroup": "claude-haiku-4-5-20251001",
    "paperUrl": "https://www.anthropic.com/news/claude-haiku-4-5",
    "codeUrl": "",
    "dataUrl": "https://huggingface.co/datasets/arcprize/arc_agi_v1_public_eval/tree/main/claude-haiku-4-5-20251001-thinking-32k",
    "featured": null
  },
  {
    "id": "gemini-3-pro-preview",
    "displayName": "Gemini 3 Pro",
    "modelReleaseDate": "2025-11-18T00:00:00.000Z",
    "providerId": "Google",
    "modelType": "CoT",
    "modelGroup": "gemini-3-pro",
    "paperUrl": null,
    "codeUrl": null,
    "dataUrl": null,
    "featured": null
  },
  {
    "id": "gemini-3-deep-think-preview",
    "displayName": "Gemini 3 Deep Think (Preview) ²",
    "modelReleaseDate": "2026-02-12T16:13:00.000Z",
    "providerId": "Google",
    "modelType": "CoT",
    "modelGroup": "gemini-3-deep-think-preview",
    "paperUrl": null,
    "codeUrl": null,
    "dataUrl": null,
    "featured": null
  },
  {
    "id": "claude-opus-4-5-20251101-thinking-none",
    "displayName": "Opus 4.5 (Thinking, None)",
    "modelReleaseDate": "2025-11-24T00:00:00.000Z",
    "providerId": "Anthropic",
    "modelType": "CoT",
    "modelGroup": "opus-4-5-thinking",
    "paperUrl": "",
    "codeUrl": "",
    "dataUrl": "",
    "featured": null
  },
  {
    "id": "claude-opus-4-5-20251101-thinking-1k",
    "displayName": "Opus 4.5 (Thinking, 1K)",
    "modelReleaseDate": "2025-11-24T00:00:00.000Z",
    "providerId": "Anthropic",
    "modelType": "CoT",
    "modelGroup": "opus-4-5-thinking",
    "paperUrl": "",
    "codeUrl": "",
    "dataUrl": "",
    "featured": null
  },
  {
    "id": "claude-opus-4-5-20251101-thinking-8k",
    "displayName": "Opus 4.5 (Thinking, 8K)",
    "modelReleaseDate": "2025-11-24T00:00:00.000Z",
    "providerId": "Anthropic",
    "modelType": "CoT",
    "modelGroup": "opus-4-5-thinking",
    "paperUrl": "",
    "codeUrl": "",
    "dataUrl": "",
    "featured": null
  },
  {
    "id": "claude-opus-4-5-20251101-thinking-16k",
    "displayName": "Opus 4.5 (Thinking, 16K)",
    "modelReleaseDate": "2025-11-24T00:00:00.000Z",
    "providerId": "Anthropic",
    "modelType": "CoT",
    "modelGroup": "opus-4-5-thinking",
    "paperUrl": "",
    "codeUrl": "",
    "dataUrl": "",
    "featured": null
  },
  {
    "id": "claude-opus-4-5-20251101-thinking-32k",
    "displayName": "Opus 4.5 (Thinking, 32K)",
    "modelReleaseDate": "2025-11-24T00:00:00.000Z",
    "providerId": "Anthropic",
    "modelType": "CoT",
    "modelGroup": "opus-4-5-thinking",
    "paperUrl": "",
    "codeUrl": "",
    "dataUrl": "",
    "featured": null
  },
  {
    "id": "claude-opus-4-5-20251101-thinking-64k",
    "displayName": "Opus 4.5 (Thinking, 64K)",
    "modelReleaseDate": "2025-11-24T00:00:00.000Z",
    "providerId": "Anthropic",
    "modelType": "CoT",
    "modelGroup": "opus-4-5-thinking",
    "paperUrl": "",
    "codeUrl": "",
    "dataUrl": "",
    "featured": null
  },
  {
    "id": "gpt-5-2-2025-12-11-thinking-none",
    "displayName": "GPT-5.2",
    "modelReleaseDate": "2025-12-11T00:00:00.000Z",
    "providerId": "OpenAI",
    "modelType": "Base LLM",
    "modelGroup": "",
    "paperUrl": "",
    "codeUrl": "",
    "dataUrl": "",
    "featured": null
  },
  {
    "id": "gpt-5-2-2025-12-11-thinking-low",
    "displayName": "GPT-5.2 (Low)",
    "modelReleaseDate": "2025-12-11T00:00:00.000Z",
    "providerId": "OpenAI",
    "modelType": "CoT",
    "modelGroup": "gpt-5-2-2025-12-11-thinking",
    "paperUrl": "",
    "codeUrl": "",
    "dataUrl": "",
    "featured": null
  },
  {
    "id": "gpt-5-2-2025-12-11-thinking-medium",
    "displayName": "GPT-5.2 (Medium)",
    "modelReleaseDate": "2025-12-11T00:00:00.000Z",
    "providerId": "OpenAI",
    "modelType": "CoT",
    "modelGroup": "gpt-5-2-2025-12-11-thinking",
    "paperUrl": "",
    "codeUrl": "",
    "dataUrl": "",
    "featured": null
  },
  {
    "id": "gpt-5-2-2025-12-11-thinking-high",
    "displayName": "GPT-5.2 (High)",
    "modelReleaseDate": "2025-12-11T00:00:00.000Z",
    "providerId": "OpenAI",
    "modelType": "CoT",
    "modelGroup": "gpt-5-2-2025-12-11-thinking",
    "paperUrl": "",
    "codeUrl": "",
    "dataUrl": "",
    "featured": null
  },
  {
    "id": "gpt-5-2-2025-12-11-thinking-xhigh",
    "displayName": "GPT-5.2 (XHigh)",
    "modelReleaseDate": "2025-12-11T00:00:00.000Z",
    "providerId": "OpenAI",
    "modelType": "CoT",
    "modelGroup": "gpt-5-2-2025-12-11-thinking",
    "paperUrl": "",
    "codeUrl": "",
    "dataUrl": "",
    "featured": null
  },
  {
    "id": "gpt-5-2-pro-2025-12-11-medium",
    "displayName": "GPT-5.2 Pro (Medium)",
    "modelReleaseDate": "2025-12-11T00:00:00.000Z",
    "providerId": "OpenAI",
    "modelType": "CoT",
    "modelGroup": "gpt-5-2-pro-2025-12-11-thinking",
    "paperUrl": "",
    "codeUrl": "",
    "dataUrl": "",
    "featured": null
  },
  {
    "id": "gpt-5-2-pro-2025-12-11-high",
    "displayName": "GPT-5.2 Pro (High)",
    "modelReleaseDate": "2025-12-11T00:00:00.000Z",
    "providerId": "OpenAI",
    "modelType": "CoT",
    "modelGroup": "gpt-5-2-pro-2025-12-11-thinking",
    "paperUrl": "",
    "codeUrl": "",
    "dataUrl": "",
    "featured": null
  },
  {
    "id": "gpt-5-2-pro-2025-12-11-xhigh",
    "displayName": "GPT-5.2 Pro (XHigh)",
    "modelReleaseDate": "2025-12-11T00:00:00.000Z",
    "providerId": "OpenAI",
    "modelType": "CoT",
    "modelGroup": "gpt-5-2-pro-2025-12-11-thinking",
    "paperUrl": "",
    "codeUrl": "",
    "dataUrl": "",
    "featured": null
  },
  {
    "id": "gemini-3-flash-preview-thinking-minimal",
    "displayName": "Gemini 3 Flash Preview (Minimal)",
    "modelReleaseDate": "2025-12-17T00:00:00.000Z",
    "providerId": "Google",
    "modelType": "CoT",
    "modelGroup": "gemini-3-flash-preview-thinking",
    "paperUrl": "",
    "codeUrl": "",
    "dataUrl": "",
    "featured": null
  },
  {
    "id": "gemini-3-flash-preview-thinking-low",
    "displayName": "Gemini 3 Flash Preview (Low)",
    "modelReleaseDate": "2025-12-17T00:00:00.000Z",
    "providerId": "Google",
    "modelType": "CoT",
    "modelGroup": "gemini-3-flash-preview-thinking",
    "paperUrl": "",
    "codeUrl": "",
    "dataUrl": "",
    "featured": null
  },
  {
    "id": "gemini-3-flash-preview-thinking-medium",
    "displayName": "Gemini 3 Flash Preview (Medium)",
    "modelReleaseDate": "2025-12-17T00:00:00.000Z",
    "providerId": "Google",
    "modelType": "CoT",
    "modelGroup": "gemini-3-flash-preview-thinking",
    "paperUrl": "",
    "codeUrl": "",
    "dataUrl": "",
    "featured": null
  },
  {
    "id": "gemini-3-flash-preview-thinking-high",
    "displayName": "Gemini 3 Flash Preview (High)",
    "modelReleaseDate": "2025-12-17T00:00:00.000Z",
    "providerId": "Google",
    "modelType": "CoT",
    "modelGroup": "gemini-3-flash-preview-thinking",
    "paperUrl": "",
    "codeUrl": "",
    "dataUrl": "",
    "featured": null
  },
  {
    "id": "grok-4.20-beta-0309b-reasoning",
    "displayName": "Grok 4.20 (Reasoning)",
    "modelReleaseDate": "2026-03-09T00:00:00.000Z",
    "providerId": "xAI",
    "modelType": "CoT",
    "modelGroup": "grok-4.20-beta-0309b-reasoning",
    "paperUrl": "",
    "codeUrl": "",
    "dataUrl": null,
    "featured": null
  },
  {
    "id": "johan_land_gpt_5_2_refine",
    "displayName": "GPT-5.2 (Refine.)",
    "modelReleaseDate": "2026-02-03T00:00:00.000Z",
    "providerId": "johan_land",
    "modelType": "Refinement",
    "modelGroup": "",
    "paperUrl": "https://arxiv.org/abs/2606.31543",
    "codeUrl": "https://www.kaggle.com/code/johanland/johan-land-solver-v7-public",
    "dataUrl": "",
    "featured": null
  },
  {
    "id": "claude-opus-4-6-thinking-120K-low",
    "displayName": "Claude Opus 4.6 (120K, Low)",
    "modelReleaseDate": "2026-02-05T00:00:00.000Z",
    "providerId": "Anthropic",
    "modelType": "CoT",
    "modelGroup": "claude-opus-4-6-thinking",
    "paperUrl": "",
    "codeUrl": "",
    "dataUrl": "",
    "featured": null,
    "notes": "Evaluation conducted with a 120K thinking budget and a maximum context window of 128K tokens and 'low' output effort.",
    "metadata": {
      "modelReleaseDate": {
        "verified": true,
        "verifiedAt": "2026-02-20T14:36:22.826452+00:00",
        "confidence": "high",
        "sourceUrl": "https://www.anthropic.com/news/claude-opus-4-6",
        "notes": "Official Anthropic announcement confirms Claude Opus 4.6 was released on February 5, 2026. The provided date matches the official release announcement date exactly."
      }
    }
  },
  {
    "id": "claude-opus-4-6-thinking-120K-medium",
    "displayName": "Claude Opus 4.6 (120K, Medium)",
    "modelReleaseDate": "2026-02-05T00:00:00.000Z",
    "providerId": "Anthropic",
    "modelType": "CoT",
    "modelGroup": "claude-opus-4-6-thinking",
    "paperUrl": "",
    "codeUrl": "",
    "dataUrl": "",
    "featured": null,
    "notes": "Evaluation conducted with a 120K thinking budget and a maximum context window of 128K tokens and 'medium' output effort.",
    "metadata": {
      "modelReleaseDate": {
        "verified": true,
        "verifiedAt": "2026-02-20T14:36:22.826452+00:00",
        "confidence": "high",
        "sourceUrl": "https://www.anthropic.com/news/claude-opus-4-6",
        "notes": "Official Anthropic announcement confirms Claude Opus 4.6 was released on February 5, 2026. The provided date matches the official release announcement date exactly."
      }
    }
  },
  {
    "id": "claude-opus-4-6-thinking-120K-high",
    "displayName": "Claude Opus 4.6 (120K, High)",
    "modelReleaseDate": "2026-02-05T00:00:00.000Z",
    "providerId": "Anthropic",
    "modelType": "CoT",
    "modelGroup": "claude-opus-4-6-thinking",
    "paperUrl": "",
    "codeUrl": "",
    "dataUrl": "",
    "featured": null,
    "notes": "Evaluation conducted with a 120K thinking budget and a maximum context window of 128K tokens and 'high' output effort.",
    "metadata": {
      "modelReleaseDate": {
        "verified": true,
        "verifiedAt": "2026-02-20T14:36:22.826452+00:00",
        "confidence": "high",
        "sourceUrl": "https://www.anthropic.com/news/claude-opus-4-6",
        "notes": "Official Anthropic announcement confirms Claude Opus 4.6 was released on February 5, 2026. The provided date matches the official release announcement date exactly."
      }
    }
  },
  {
    "id": "claude-opus-4-6-thinking-120K-max",
    "displayName": "Claude Opus 4.6 (120K, Max)",
    "modelReleaseDate": "2026-02-05T00:00:00.000Z",
    "providerId": "Anthropic",
    "modelType": "CoT",
    "modelGroup": "claude-opus-4-6-thinking",
    "paperUrl": "",
    "codeUrl": "",
    "dataUrl": "",
    "featured": null,
    "notes": "Evaluation conducted with a 120K thinking budget and a maximum context window of 128K tokens and 'max' output effort.",
    "metadata": {
      "modelReleaseDate": {
        "verified": true,
        "verifiedAt": "2026-02-20T14:36:22.826452+00:00",
        "confidence": "high",
        "sourceUrl": "https://www.anthropic.com/news/claude-opus-4-6",
        "notes": "Official Anthropic announcement confirms Claude Opus 4.6 was released on February 5, 2026. The provided date matches the official release announcement date exactly."
      }
    }
  },
  {
    "id": "anthropic-opus-4-8-low",
    "displayName": "Claude Opus 4.8 (Low)",
    "modelReleaseDate": "2026-06-01T00:00:00.000Z",
    "providerId": "Anthropic",
    "modelType": "CoT",
    "modelGroup": "anthropic-opus-4-8",
    "paperUrl": "",
    "codeUrl": "",
    "dataUrl": "",
    "featured": null,
    "notes": "Evaluation conducted with 'low' output effort."
  },
  {
    "id": "anthropic-opus-4-8-medium",
    "displayName": "Claude Opus 4.8 (Medium)",
    "modelReleaseDate": "2026-06-01T00:00:00.000Z",
    "providerId": "Anthropic",
    "modelType": "CoT",
    "modelGroup": "anthropic-opus-4-8",
    "paperUrl": "",
    "codeUrl": "",
    "dataUrl": "",
    "featured": null,
    "notes": "Evaluation conducted with 'medium' output effort."
  },
  {
    "id": "anthropic-opus-4-8-high",
    "displayName": "Claude Opus 4.8 (High)",
    "modelReleaseDate": "2026-06-01T00:00:00.000Z",
    "providerId": "Anthropic",
    "modelType": "CoT",
    "modelGroup": "anthropic-opus-4-8",
    "paperUrl": "",
    "codeUrl": "",
    "dataUrl": "",
    "featured": true,
    "notes": "Evaluation conducted with 'high' output effort."
  },
  {
    "id": "anthropic-opus-4-8-max",
    "displayName": "Claude Opus 4.8 (Max)",
    "modelReleaseDate": "2026-06-01T00:00:00.000Z",
    "providerId": "Anthropic",
    "modelType": "CoT",
    "modelGroup": "anthropic-opus-4-8",
    "paperUrl": "",
    "codeUrl": "",
    "dataUrl": "",
    "featured": null,
    "notes": "Evaluation conducted with 'max' output effort."
  },
  {
    "id": "gemini_3_deep_think_022026",
    "displayName": "Gemini 3 Deep Think (2/26)",
    "modelReleaseDate": "2026-02-12T16:13:00.000Z",
    "providerId": "Google",
    "modelType": "CoT",
    "modelGroup": "",
    "paperUrl": "",
    "codeUrl": "",
    "dataUrl": "",
    "featured": null,
    "notes": "",
    "metadata": {
      "modelReleaseDate": {
        "verified": true,
        "verifiedAt": "2026-02-20T14:36:21.114898+00:00",
        "confidence": "high",
        "sourceUrl": "https://blog.google/innovation-and-ai/models-and-research/gemini-models/gemini-3-deep-think/",
        "notes": "Date corrected from 2026-02-20T00:00:00.000Z to 2026-02-12T16:13:00.000Z. The official Google blog post announcing Gemini 3 Deep Think was published on February 12, 2026 at 16:13 UTC. The recorded date of 2026-02-20 is 8 days later than the actual release date.",
        "previousValue": "2026-02-20T00:00:00.000Z"
      }
    }
  },
  {
    "id": "claude_sonnet_4_6_max",
    "displayName": "Claude Sonnet 4.6 (Max)",
    "modelReleaseDate": "2026-02-17T00:00:00.000Z",
    "providerId": "Anthropic",
    "modelType": "CoT",
    "modelGroup": "claude-sonnet-4-6",
    "paperUrl": "",
    "codeUrl": "",
    "dataUrl": null,
    "featured": null,
    "notes": "Evaluation conducted with a 120K thinking budget and a maximum context window of 128K tokens and 'max' output effort.",
    "metadata": {
      "modelReleaseDate": {
        "verified": true,
        "verifiedAt": "2026-02-20T14:36:33.270495+00:00",
        "confidence": "high",
        "sourceUrl": "https://www.anthropic.com/news/claude-sonnet-4-6",
        "notes": "Claude Sonnet 4.6 was officially announced by Anthropic on February 17, 2026, as confirmed by the official announcement blog post. The recorded date of 2026-02-17T00:00:00.000Z matches the official release announcement date."
      }
    }
  },
  {
    "id": "claude_sonnet_4_6_high",
    "displayName": "Claude Sonnet 4.6 (High)",
    "modelReleaseDate": "2026-02-17T00:00:00.000Z",
    "providerId": "Anthropic",
    "modelType": "CoT",
    "modelGroup": "claude-sonnet-4-6",
    "paperUrl": "",
    "codeUrl": "",
    "dataUrl": null,
    "featured": null,
    "notes": "Evaluation conducted with a 120K thinking budget and a maximum context window of 128K tokens and 'high' output effort.",
    "metadata": {
      "modelReleaseDate": {
        "verified": true,
        "verifiedAt": "2026-02-20T14:36:33.270495+00:00",
        "confidence": "high",
        "sourceUrl": "https://www.anthropic.com/news/claude-sonnet-4-6",
        "notes": "Claude Sonnet 4.6 was officially announced by Anthropic on February 17, 2026, as confirmed by the official announcement blog post. The recorded date of 2026-02-17T00:00:00.000Z matches the official release announcement date."
      }
    }
  },
  {
    "id": "gemini-3-1-pro-preview",
    "displayName": "Gemini 3.1 Pro (Preview)",
    "modelReleaseDate": "2026-02-19T00:00:00.000Z",
    "providerId": "Google",
    "modelType": "CoT",
    "modelGroup": "",
    "paperUrl": "https://blog.google/innovation-and-ai/models-and-research/gemini-models/gemini-3-1-pro/",
    "codeUrl": "",
    "dataUrl": null,
    "featured": null,
    "metadata": {
      "modelReleaseDate": {
        "verified": true,
        "verifiedAt": "2026-02-20T14:41:53.556515+00:00",
        "confidence": "high",
        "sourceUrl": "https://blog.google/innovation-and-ai/models-and-research/gemini-models/gemini-3-1-pro/",
        "notes": "Official Google blog announcement confirms Gemini 3.1 Pro was announced on February 19, 2026. The model was released in preview across multiple platforms including Gemini API, Vertex AI, and the Gemini app."
      }
    }
  },
  {
    "id": "kimi-k2.5",
    "displayName": "Kimi K2.5",
    "modelReleaseDate": "2026-01-27T00:00:00.000Z",
    "providerId": "Moonshot AI",
    "modelType": "CoT",
    "modelGroup": "",
    "paperUrl": "",
    "codeUrl": "",
    "dataUrl": "https://huggingface.co/datasets/arcprize/arc_agi_v1_public_eval/tree/main/kimi-k2.5",
    "featured": null,
    "metadata": {
      "modelReleaseDate": {
        "verified": true,
        "verifiedAt": "2026-02-20T14:36:20.108709+00:00",
        "confidence": "high",
        "sourceUrl": "https://techcrunch.com/2026/01/27/chinas-moonshot-releases-a-new-open-source-model-kimi-k2-5-and-a-coding-agent/",
        "notes": "Date corrected from 2025-01-27T00:00:00.000Z to 2026-01-27T00:00:00.000Z. Kimi K2.5 was officially released by Moonshot AI on January 27, 2026, not 2025. Multiple official sources and news outlets confirm the 2026 release date. The model is an open-weight multimodal agentic model with 1 trillion parameters featuring Agent Swarm capabilities.",
        "previousValue": "2025-01-27T00:00:00.000Z"
      }
    }
  },
  {
    "id": "minimax-m2.5",
    "displayName": "Minimax M2.5",
    "modelReleaseDate": "2026-02-12T00:00:00.000Z",
    "providerId": "Minimax",
    "modelType": "CoT",
    "modelGroup": "",
    "paperUrl": "",
    "codeUrl": "",
    "dataUrl": "https://huggingface.co/datasets/arcprize/arc_agi_v1_public_eval/tree/main/minimax-m2.5",
    "featured": null
  },
  {
    "id": "deepseek-v3.2",
    "displayName": "Deepseek V3.2",
    "modelReleaseDate": "2025-12-01T00:00:00.000Z",
    "providerId": "DeepSeek",
    "modelType": "Base LLM",
    "modelGroup": "",
    "paperUrl": "",
    "codeUrl": "",
    "dataUrl": null,
    "featured": null
  },
  {
    "id": "glm-5",
    "displayName": "GLM-5",
    "modelReleaseDate": "2026-02-20T00:00:00.000Z",
    "providerId": "Z.ai",
    "modelType": "CoT",
    "modelGroup": "",
    "paperUrl": "",
    "codeUrl": "",
    "dataUrl": "https://huggingface.co/datasets/arcprize/arc_agi_v1_public_eval/tree/main/glm-5",
    "featured": null
  },
  {
    "id": "glm-5.2",
    "displayName": "GLM-5.2",
    "modelReleaseDate": "2026-06-13T00:00:00.000Z",
    "providerId": "Z.ai",
    "modelType": "CoT",
    "modelGroup": "",
    "paperUrl": "",
    "codeUrl": "",
    "dataUrl": "https://docs.z.ai/guides/llm/glm-5.2",
    "featured": null
  },
  {
    "id": "gpt-5-4-low",
    "displayName": "GPT-5.4 (Low)",
    "modelReleaseDate": "2026-03-04T00:00:00.000Z",
    "providerId": "OpenAI",
    "modelType": "CoT",
    "modelGroup": "gpt-5-4",
    "paperUrl": "",
    "codeUrl": "",
    "dataUrl": "",
    "featured": null
  },
  {
    "id": "gpt-5-4-medium",
    "displayName": "GPT-5.4 (Medium)",
    "modelReleaseDate": "2026-03-04T00:00:00.000Z",
    "providerId": "OpenAI",
    "modelType": "CoT",
    "modelGroup": "gpt-5-4",
    "paperUrl": "",
    "codeUrl": "",
    "dataUrl": "",
    "featured": null
  },
  {
    "id": "gpt-5-4-high",
    "displayName": "GPT-5.4 (High)",
    "modelReleaseDate": "2026-03-04T00:00:00.000Z",
    "providerId": "OpenAI",
    "modelType": "CoT",
    "modelGroup": "gpt-5-4",
    "paperUrl": "",
    "codeUrl": "",
    "dataUrl": "",
    "featured": null
  },
  {
    "id": "gpt-5-4-xhigh",
    "displayName": "GPT-5.4 (XHigh)",
    "modelReleaseDate": "2026-03-04T00:00:00.000Z",
    "providerId": "OpenAI",
    "modelType": "CoT",
    "modelGroup": "gpt-5-4",
    "paperUrl": "",
    "codeUrl": "",
    "dataUrl": "",
    "featured": null
  },
  {
    "id": "gpt-5-4-pro-xhigh",
    "displayName": "GPT-5.4 Pro (XHigh)",
    "modelReleaseDate": "2026-03-04T00:00:00.000Z",
    "providerId": "OpenAI",
    "modelType": "CoT",
    "modelGroup": "gpt-5-4-pro",
    "paperUrl": "",
    "codeUrl": "",
    "dataUrl": "",
    "featured": null
  },
  {
    "id": "gpt-5-4-mini-xhigh",
    "displayName": "GPT-5.4 Mini (XHigh)",
    "modelReleaseDate": "2026-03-17T00:00:00.000Z",
    "providerId": "OpenAI",
    "modelType": "CoT",
    "modelGroup": "gpt-5-4-mini-reasoning",
    "paperUrl": "",
    "codeUrl": "",
    "dataUrl": "",
    "featured": null
  },
  {
    "id": "gpt-5-4-mini-high",
    "displayName": "GPT-5.4 Mini (High)",
    "modelReleaseDate": "2026-03-17T00:00:00.000Z",
    "providerId": "OpenAI",
    "modelType": "CoT",
    "modelGroup": "gpt-5-4-mini-reasoning",
    "paperUrl": "",
    "codeUrl": "",
    "dataUrl": "",
    "featured": null
  },
  {
    "id": "gpt-5-4-mini-medium",
    "displayName": "GPT-5.4 Mini (Medium)",
    "modelReleaseDate": "2026-03-17T00:00:00.000Z",
    "providerId": "OpenAI",
    "modelType": "CoT",
    "modelGroup": "gpt-5-4-mini-reasoning",
    "paperUrl": "",
    "codeUrl": "",
    "dataUrl": "",
    "featured": null
  },
  {
    "id": "gpt-5-4-mini-low",
    "displayName": "GPT-5.4 Mini (Low)",
    "modelReleaseDate": "2026-03-17T00:00:00.000Z",
    "providerId": "OpenAI",
    "modelType": "CoT",
    "modelGroup": "gpt-5-4-mini-reasoning",
    "paperUrl": "",
    "codeUrl": "",
    "dataUrl": "",
    "featured": null
  },
  {
    "id": "gpt-5-4-nano-xhigh",
    "displayName": "GPT-5.4 Nano (XHigh)",
    "modelReleaseDate": "2026-03-17T00:00:00.000Z",
    "providerId": "OpenAI",
    "modelType": "CoT",
    "modelGroup": "gpt-5-4-nano-reasoning",
    "paperUrl": "",
    "codeUrl": "",
    "dataUrl": "",
    "featured": null
  },
  {
    "id": "gpt-5-4-nano-high",
    "displayName": "GPT-5.4 Nano (High)",
    "modelReleaseDate": "2026-03-17T00:00:00.000Z",
    "providerId": "OpenAI",
    "modelType": "CoT",
    "modelGroup": "gpt-5-4-nano-reasoning",
    "paperUrl": "",
    "codeUrl": "",
    "dataUrl": "",
    "featured": null
  },
  {
    "id": "gpt-5-4-nano-medium",
    "displayName": "GPT-5.4 Nano (Medium)",
    "modelReleaseDate": "2026-03-17T00:00:00.000Z",
    "providerId": "OpenAI",
    "modelType": "CoT",
    "modelGroup": "gpt-5-4-nano-reasoning",
    "paperUrl": "",
    "codeUrl": "",
    "dataUrl": "",
    "featured": null
  },
  {
    "id": "gpt-5-4-nano-low",
    "displayName": "GPT-5.4 Nano (Low)",
    "modelReleaseDate": "2026-03-17T00:00:00.000Z",
    "providerId": "OpenAI",
    "modelType": "CoT",
    "modelGroup": "gpt-5-4-nano-reasoning",
    "paperUrl": "",
    "codeUrl": "",
    "dataUrl": "",
    "featured": null
  },
  {
    "id": "anthropic-opus-4-6-max-effort",
    "displayName": "Anthropic Opus 4.6 (Max)",
    "modelReleaseDate": "2025-12-17T00:00:00.000Z",
    "providerId": "Anthropic",
    "modelType": "CoT",
    "modelGroup": "anthropic-opus-4-6-max-effort",
    "paperUrl": "",
    "codeUrl": "",
    "dataUrl": "",
    "featured": null
  },
  {
    "id": "openai-gpt-5-4-2026-03-05-high",
    "displayName": "GPT-5.4 (High)",
    "modelReleaseDate": "2026-03-05T00:00:00.000Z",
    "providerId": "OpenAI",
    "modelType": "CoT",
    "modelGroup": "openai-gpt-5-4-2026-03-05-thinking",
    "paperUrl": "",
    "codeUrl": "",
    "dataUrl": "",
    "featured": null
  },
  {
    "id": "xai-grok-4-20-beta-0309-reasoning",
    "displayName": "Grok 4.20 (Beta Reasoning)",
    "modelReleaseDate": "2026-03-05T00:00:00.000Z",
    "providerId": "xAI",
    "modelType": "CoT",
    "modelGroup": "xai-grok-4-20-beta-0309-reasoning",
    "paperUrl": "",
    "codeUrl": "",
    "dataUrl": "",
    "featured": true
  },
  {
    "id": "google-gemini-3-1-pro-preview",
    "displayName": "Gemini 3.1 Pro (Preview)",
    "modelReleaseDate": "2026-03-05T00:00:00.000Z",
    "providerId": "Google",
    "modelType": "CoT",
    "modelGroup": "google-gemini-3-1-pro-preview",
    "paperUrl": "",
    "codeUrl": "",
    "dataUrl": "",
    "featured": true
  },
  {
    "id": "claude-4-7-low",
    "displayName": "Claude 4.7 (Low)",
    "modelReleaseDate": "2026-04-16T00:00:00.000Z",
    "providerId": "Anthropic",
    "modelType": "CoT",
    "modelGroup": "claude-4-7",
    "paperUrl": "",
    "codeUrl": "",
    "dataUrl": "",
    "featured": null
  },
  {
    "id": "claude-4-7-medium",
    "displayName": "Claude 4.7 (Medium)",
    "modelReleaseDate": "2026-04-16T00:00:00.000Z",
    "providerId": "Anthropic",
    "modelType": "CoT",
    "modelGroup": "claude-4-7",
    "paperUrl": "",
    "codeUrl": "",
    "dataUrl": "",
    "featured": null
  },
  {
    "id": "claude-4-7-high",
    "displayName": "Claude 4.7 (High)",
    "modelReleaseDate": "2026-04-16T00:00:00.000Z",
    "providerId": "Anthropic",
    "modelType": "CoT",
    "modelGroup": "claude-4-7",
    "paperUrl": "",
    "codeUrl": "",
    "dataUrl": "",
    "featured": null
  },
  {
    "id": "claude-4-7-max",
    "displayName": "Claude 4.7 (Max)",
    "modelReleaseDate": "2026-04-16T00:00:00.000Z",
    "providerId": "Anthropic",
    "modelType": "CoT",
    "modelGroup": "claude-4-7",
    "paperUrl": "",
    "codeUrl": "",
    "dataUrl": "",
    "featured": null
  },
  {
    "id": "gpt-5-5-2026-04-22-thinking-low",
    "displayName": "GPT-5.5 (Low)",
    "modelReleaseDate": "2026-04-22T00:00:00.000Z",
    "providerId": "OpenAI",
    "modelType": "CoT",
    "modelGroup": "gpt-5-5-2026-04-22-thinking",
    "paperUrl": "",
    "codeUrl": "",
    "dataUrl": "",
    "featured": null
  },
  {
    "id": "gpt-5-5-2026-04-22-thinking-medium",
    "displayName": "GPT-5.5 (Medium)",
    "modelReleaseDate": "2026-04-22T00:00:00.000Z",
    "providerId": "OpenAI",
    "modelType": "CoT",
    "modelGroup": "gpt-5-5-2026-04-22-thinking",
    "paperUrl": "",
    "codeUrl": "",
    "dataUrl": "",
    "featured": null
  },
  {
    "id": "gpt-5-5-2026-04-22-thinking-high",
    "displayName": "GPT-5.5 (High)",
    "modelReleaseDate": "2026-04-22T00:00:00.000Z",
    "providerId": "OpenAI",
    "modelType": "CoT",
    "modelGroup": "gpt-5-5-2026-04-22-thinking",
    "paperUrl": "",
    "codeUrl": "",
    "dataUrl": "",
    "featured": null
  },
  {
    "id": "gpt-5-5-2026-04-22-thinking-xhigh",
    "displayName": "GPT-5.5 (XHigh)",
    "modelReleaseDate": "2026-04-22T00:00:00.000Z",
    "providerId": "OpenAI",
    "modelType": "CoT",
    "modelGroup": "gpt-5-5-2026-04-22-thinking",
    "paperUrl": "",
    "codeUrl": "",
    "dataUrl": "",
    "featured": null
  },
  {
    "id": "gpt-5-5-pro-2026-04-23-high",
    "displayName": "GPT-5.5 Pro (High)",
    "modelReleaseDate": "2026-04-23T00:00:00.000Z",
    "providerId": "OpenAI",
    "modelType": "CoT",
    "modelGroup": "gpt-5-5-pro-2026-04-23-thinking",
    "paperUrl": "",
    "codeUrl": "",
    "dataUrl": "",
    "featured": null
  },
  {
    "id": "gpt-5-5-pro-2026-04-23-xhigh",
    "displayName": "GPT-5.5 Pro (XHigh)",
    "modelReleaseDate": "2026-04-23T00:00:00.000Z",
    "providerId": "OpenAI",
    "modelType": "CoT",
    "modelGroup": "gpt-5-5-pro-2026-04-23-thinking",
    "paperUrl": "",
    "codeUrl": "",
    "dataUrl": "",
    "featured": null
  },
  {
    "id": "openai-gpt-5-5-2026-04-23-high",
    "displayName": "GPT-5.5 (High)",
    "modelReleaseDate": "2026-04-23T00:00:00.000Z",
    "providerId": "OpenAI",
    "modelType": "CoT",
    "modelGroup": "openai-gpt-5-5-2026-04-23-thinking",
    "paperUrl": "",
    "codeUrl": "",
    "dataUrl": "",
    "featured": true
  },
  {
    "id": "anthropic-opus-4-7-high",
    "displayName": "Opus 4.7 (High)",
    "modelReleaseDate": "2026-04-16T00:00:00.000Z",
    "providerId": "Anthropic",
    "modelType": "CoT",
    "modelGroup": "anthropic-opus-4-7",
    "paperUrl": "",
    "codeUrl": "",
    "dataUrl": "",
    "featured": null
  },
  {
    "id": "gemini-3-5-flash-minimal",
    "displayName": "Gemini 3.5 Flash (Minimal)",
    "modelReleaseDate": "2026-05-19T00:00:00.000Z",
    "providerId": "Google",
    "modelType": "CoT",
    "modelGroup": "gemini-3-5-flash",
    "paperUrl": "",
    "codeUrl": "",
    "dataUrl": "",
    "featured": null
  },
  {
    "id": "gemini-3-5-flash-high",
    "displayName": "Gemini 3.5 Flash (High)",
    "modelReleaseDate": "2026-05-19T00:00:00.000Z",
    "providerId": "Google",
    "modelType": "CoT",
    "modelGroup": "gemini-3-5-flash",
    "paperUrl": "",
    "codeUrl": "",
    "dataUrl": "",
    "featured": null
  },
  {
    "id": "openai-gpt-5-6-sol-low",
    "displayName": "GPT-5.6 Sol (Low)",
    "modelReleaseDate": "2026-07-09T00:00:00.000Z",
    "providerId": "OpenAI",
    "modelType": "CoT",
    "modelGroup": "openai-gpt-5-6-sol",
    "paperUrl": "",
    "codeUrl": "",
    "dataUrl": "",
    "featured": null
  },
  {
    "id": "openai-gpt-5-6-sol-medium",
    "displayName": "GPT-5.6 Sol (Medium)",
    "modelReleaseDate": "2026-07-09T00:00:00.000Z",
    "providerId": "OpenAI",
    "modelType": "CoT",
    "modelGroup": "openai-gpt-5-6-sol",
    "paperUrl": "",
    "codeUrl": "",
    "dataUrl": "",
    "featured": null
  },
  {
    "id": "openai-gpt-5-6-sol-high",
    "displayName": "GPT-5.6 Sol (High)",
    "modelReleaseDate": "2026-07-09T00:00:00.000Z",
    "providerId": "OpenAI",
    "modelType": "CoT",
    "modelGroup": "openai-gpt-5-6-sol",
    "paperUrl": "",
    "codeUrl": "",
    "dataUrl": "",
    "featured": null
  },
  {
    "id": "openai-gpt-5-6-sol-xhigh",
    "displayName": "GPT-5.6 Sol (XHigh)",
    "modelReleaseDate": "2026-07-09T00:00:00.000Z",
    "providerId": "OpenAI",
    "modelType": "CoT",
    "modelGroup": "openai-gpt-5-6-sol",
    "paperUrl": "",
    "codeUrl": "",
    "dataUrl": "",
    "featured": null
  },
  {
    "id": "openai-gpt-5-6-sol-max",
    "displayName": "GPT-5.6 Sol (Max)",
    "modelReleaseDate": "2026-07-09T00:00:00.000Z",
    "providerId": "OpenAI",
    "modelType": "CoT",
    "modelGroup": "openai-gpt-5-6-sol",
    "paperUrl": "",
    "codeUrl": "",
    "dataUrl": "",
    "featured": null
  },
  {
    "id": "openai-gpt-5-6-terra-low",
    "displayName": "GPT-5.6 Terra (Low)",
    "modelReleaseDate": "2026-07-09T00:00:00.000Z",
    "providerId": "OpenAI",
    "modelType": "CoT",
    "modelGroup": "openai-gpt-5-6-terra",
    "paperUrl": "",
    "codeUrl": "",
    "dataUrl": "",
    "featured": null
  },
  {
    "id": "openai-gpt-5-6-terra-medium",
    "displayName": "GPT-5.6 Terra (Medium)",
    "modelReleaseDate": "2026-07-09T00:00:00.000Z",
    "providerId": "OpenAI",
    "modelType": "CoT",
    "modelGroup": "openai-gpt-5-6-terra",
    "paperUrl": "",
    "codeUrl": "",
    "dataUrl": "",
    "featured": null
  },
  {
    "id": "openai-gpt-5-6-terra-high",
    "displayName": "GPT-5.6 Terra (High)",
    "modelReleaseDate": "2026-07-09T00:00:00.000Z",
    "providerId": "OpenAI",
    "modelType": "CoT",
    "modelGroup": "openai-gpt-5-6-terra",
    "paperUrl": "",
    "codeUrl": "",
    "dataUrl": "",
    "featured": null
  },
  {
    "id": "openai-gpt-5-6-terra-xhigh",
    "displayName": "GPT-5.6 Terra (XHigh)",
    "modelReleaseDate": "2026-07-09T00:00:00.000Z",
    "providerId": "OpenAI",
    "modelType": "CoT",
    "modelGroup": "openai-gpt-5-6-terra",
    "paperUrl": "",
    "codeUrl": "",
    "dataUrl": "",
    "featured": null
  },
  {
    "id": "openai-gpt-5-6-terra-max",
    "displayName": "GPT-5.6 Terra (Max)",
    "modelReleaseDate": "2026-07-09T00:00:00.000Z",
    "providerId": "OpenAI",
    "modelType": "CoT",
    "modelGroup": "openai-gpt-5-6-terra",
    "paperUrl": "",
    "codeUrl": "",
    "dataUrl": "",
    "featured": null
  },
  {
    "id": "openai-gpt-5-6-luna-low",
    "displayName": "GPT-5.6 Luna (Low)",
    "modelReleaseDate": "2026-07-09T00:00:00.000Z",
    "providerId": "OpenAI",
    "modelType": "CoT",
    "modelGroup": "openai-gpt-5-6-luna",
    "paperUrl": "",
    "codeUrl": "",
    "dataUrl": "",
    "featured": null
  },
  {
    "id": "openai-gpt-5-6-luna-medium",
    "displayName": "GPT-5.6 Luna (Medium)",
    "modelReleaseDate": "2026-07-09T00:00:00.000Z",
    "providerId": "OpenAI",
    "modelType": "CoT",
    "modelGroup": "openai-gpt-5-6-luna",
    "paperUrl": "",
    "codeUrl": "",
    "dataUrl": "",
    "featured": null
  },
  {
    "id": "openai-gpt-5-6-luna-high",
    "displayName": "GPT-5.6 Luna (High)",
    "modelReleaseDate": "2026-07-09T00:00:00.000Z",
    "providerId": "OpenAI",
    "modelType": "CoT",
    "modelGroup": "openai-gpt-5-6-luna",
    "paperUrl": "",
    "codeUrl": "",
    "dataUrl": "",
    "featured": null
  },
  {
    "id": "openai-gpt-5-6-luna-xhigh",
    "displayName": "GPT-5.6 Luna (XHigh)",
    "modelReleaseDate": "2026-07-09T00:00:00.000Z",
    "providerId": "OpenAI",
    "modelType": "CoT",
    "modelGroup": "openai-gpt-5-6-luna",
    "paperUrl": "",
    "codeUrl": "",
    "dataUrl": "",
    "featured": null
  },
  {
    "id": "openai-gpt-5-6-luna-max",
    "displayName": "GPT-5.6 Luna (Max)",
    "modelReleaseDate": "2026-07-09T00:00:00.000Z",
    "providerId": "OpenAI",
    "modelType": "CoT",
    "modelGroup": "openai-gpt-5-6-luna",
    "paperUrl": "",
    "codeUrl": "",
    "dataUrl": "",
    "featured": null
  },
  {
    "id": "thinky-inkling",
    "displayName": "Inkling",
    "modelReleaseDate": "2026-07-15T00:00:00.000Z",
    "providerId": "Thinking Machines",
    "modelType": "CoT",
    "modelGroup": "",
    "paperUrl": "https://thinkingmachines.ai/news/introducing-inkling/",
    "codeUrl": "",
    "dataUrl": "https://huggingface.co/thinkingmachines/inkling",
    "featured": null
  },
  {
    "id": "xai-grok-4-5-high",
    "displayName": "Grok 4.5 (High)",
    "modelReleaseDate": "2026-07-16T00:00:00.000Z",
    "providerId": "xAI",
    "modelType": "CoT",
    "modelGroup": "xai-grok-4-5",
    "paperUrl": "https://x.ai/news/grok-4-5",
    "codeUrl": "",
    "dataUrl": "",
    "featured": null
  },
  {
    "id": "xai-grok-4-5-medium",
    "displayName": "Grok 4.5 (Medium)",
    "modelReleaseDate": "2026-07-16T00:00:00.000Z",
    "providerId": "xAI",
    "modelType": "CoT",
    "modelGroup": "xai-grok-4-5",
    "paperUrl": "https://x.ai/news/grok-4-5",
    "codeUrl": "",
    "dataUrl": "",
    "featured": null
  },
  {
    "id": "xai-grok-4-5-low",
    "displayName": "Grok 4.5 (Low)",
    "modelReleaseDate": "2026-07-16T00:00:00.000Z",
    "providerId": "xAI",
    "modelType": "CoT",
    "modelGroup": "xai-grok-4-5",
    "paperUrl": "https://x.ai/news/grok-4-5",
    "codeUrl": "",
    "dataUrl": "",
    "featured": null
  },
  {
    "id": "anthropic-claude-opus-5-high",
    "displayName": "Claude Opus 5 (High)",
    "modelReleaseDate": "2026-07-24T00:00:00.000Z",
    "providerId": "Anthropic",
    "modelType": "CoT",
    "modelGroup": "anthropic-claude-opus-5",
    "paperUrl": "/results/anthropic-claude-opus-5",
    "codeUrl": "",
    "dataUrl": "",
    "featured": true,
    "notes": "Evaluation conducted with 'high' output effort."
  },
  {
    "id": "anthropic-claude-opus-5-max",
    "displayName": "Claude Opus 5 (Max)",
    "modelReleaseDate": "2026-07-24T00:00:00.000Z",
    "providerId": "Anthropic",
    "modelType": "CoT",
    "modelGroup": "anthropic-claude-opus-5",
    "paperUrl": "/results/anthropic-claude-opus-5",
    "codeUrl": "",
    "dataUrl": "",
    "featured": null,
    "notes": "Evaluation conducted with 'max' output effort."
  },
  {
    "id": "thinky-inkling-small-xhigh",
    "displayName": "Inkling Small (XHigh)",
    "modelReleaseDate": "2026-07-30T00:00:00.000Z",
    "providerId": "Thinking Machines",
    "modelType": "CoT",
    "modelGroup": "thinky-inkling-small",
    "paperUrl": "https://thinkingmachines.ai/news/inkling-small/",
    "codeUrl": "",
    "dataUrl": "",
    "featured": null,
    "notes": "Evaluation conducted with 'xhigh' reasoning effort."
  },
  {
    "id": "thinky-inkling-small-high",
    "displayName": "Inkling Small (High)",
    "modelReleaseDate": "2026-07-30T00:00:00.000Z",
    "providerId": "Thinking Machines",
    "modelType": "CoT",
    "modelGroup": "thinky-inkling-small",
    "paperUrl": "https://thinkingmachines.ai/news/inkling-small/",
    "codeUrl": "",
    "dataUrl": "",
    "featured": null,
    "notes": "Evaluation conducted with 'high' reasoning effort."
  },
  {
    "id": "thinky-inkling-small-medium",
    "displayName": "Inkling Small (Medium)",
    "modelReleaseDate": "2026-07-30T00:00:00.000Z",
    "providerId": "Thinking Machines",
    "modelType": "CoT",
    "modelGroup": "thinky-inkling-small",
    "paperUrl": "https://thinkingmachines.ai/news/inkling-small/",
    "codeUrl": "",
    "dataUrl": "",
    "featured": null,
    "notes": "Evaluation conducted with 'medium' reasoning effort."
  },
  {
    "id": "thinky-inkling-small-low",
    "displayName": "Inkling Small (Low)",
    "modelReleaseDate": "2026-07-30T00:00:00.000Z",
    "providerId": "Thinking Machines",
    "modelType": "CoT",
    "modelGroup": "thinky-inkling-small",
    "paperUrl": "https://thinkingmachines.ai/news/inkling-small/",
    "codeUrl": "",
    "dataUrl": "",
    "featured": null,
    "notes": "Evaluation conducted with 'low' reasoning effort."
  },
  {
    "id": "thinky-inkling-small-minimal",
    "displayName": "Inkling Small (Minimal)",
    "modelReleaseDate": "2026-07-30T00:00:00.000Z",
    "providerId": "Thinking Machines",
    "modelType": "CoT",
    "modelGroup": "thinky-inkling-small",
    "paperUrl": "https://thinkingmachines.ai/news/inkling-small/",
    "codeUrl": "",
    "dataUrl": "",
    "featured": null,
    "notes": "Evaluation conducted with 'minimal' reasoning effort."
  },
  {
    "id": "thinky-inkling-small-none",
    "displayName": "Inkling Small (None)",
    "modelReleaseDate": "2026-07-30T00:00:00.000Z",
    "providerId": "Thinking Machines",
    "modelType": "CoT",
    "modelGroup": "thinky-inkling-small",
    "paperUrl": "https://thinkingmachines.ai/news/inkling-small/",
    "codeUrl": "",
    "dataUrl": "",
    "featured": null,
    "notes": "Evaluation conducted with 'none' reasoning effort."
  },
  {
    "id": "moonshot-kimi-k3-max",
    "displayName": "Kimi K3 (Max)",
    "modelReleaseDate": "2026-07-16T00:00:00.000Z",
    "providerId": "Moonshot AI",
    "modelType": "CoT",
    "modelGroup": "moonshot-kimi-k3",
    "paperUrl": "https://www.kimi.com/blog/kimi-k3",
    "codeUrl": "https://huggingface.co/moonshotai/Kimi-K3",
    "dataUrl": "",
    "featured": null,
    "notes": "Evaluation conducted with 'max' reasoning effort via Baseten."
  },
  {
    "id": "moonshot-kimi-k3-high",
    "displayName": "Kimi K3 (High)",
    "modelReleaseDate": "2026-07-16T00:00:00.000Z",
    "providerId": "Moonshot AI",
    "modelType": "CoT",
    "modelGroup": "moonshot-kimi-k3",
    "paperUrl": "https://www.kimi.com/blog/kimi-k3",
    "codeUrl": "https://huggingface.co/moonshotai/Kimi-K3",
    "dataUrl": "",
    "featured": null,
    "notes": "Evaluation conducted with 'high' reasoning effort via Baseten."
  },
  {
    "id": "moonshot-kimi-k3-low",
    "displayName": "Kimi K3 (Low)",
    "modelReleaseDate": "2026-07-16T00:00:00.000Z",
    "providerId": "Moonshot AI",
    "modelType": "CoT",
    "modelGroup": "moonshot-kimi-k3",
    "paperUrl": "https://www.kimi.com/blog/kimi-k3",
    "codeUrl": "https://huggingface.co/moonshotai/Kimi-K3",
    "dataUrl": "",
    "featured": null,
    "notes": "Evaluation conducted with 'low' reasoning effort via Baseten."
  },
  {
    "id": "anthropic-claude-fable-5-max",
    "displayName": "Claude Fable 5 (Max)",
    "modelReleaseDate": "2026-06-09T00:00:00.000Z",
    "providerId": "Anthropic",
    "modelType": "CoT",
    "modelGroup": "anthropic-claude-fable-5",
    "paperUrl": "",
    "dataUrl": "",
    "featured": null,
    "notes": "Evaluation conducted with 'max' reasoning effort."
  },
  {
    "id": "anthropic-claude-fable-5-xhigh",
    "displayName": "Claude Fable 5 (XHigh)",
    "modelReleaseDate": "2026-06-09T00:00:00.000Z",
    "providerId": "Anthropic",
    "modelType": "CoT",
    "modelGroup": "anthropic-claude-fable-5",
    "paperUrl": "",
    "dataUrl": "",
    "featured": null,
    "notes": "Evaluation conducted with 'xhigh' reasoning effort."
  },
  {
    "id": "anthropic-claude-fable-5-high",
    "displayName": "Claude Fable 5 (High)",
    "modelReleaseDate": "2026-06-09T00:00:00.000Z",
    "providerId": "Anthropic",
    "modelType": "CoT",
    "modelGroup": "anthropic-claude-fable-5",
    "paperUrl": "",
    "dataUrl": "",
    "featured": null,
    "notes": "Evaluation conducted with 'high' reasoning effort."
  },
  {
    "id": "anthropic-claude-fable-5-medium",
    "displayName": "Claude Fable 5 (Medium)",
    "modelReleaseDate": "2026-06-09T00:00:00.000Z",
    "providerId": "Anthropic",
    "modelType": "CoT",
    "modelGroup": "anthropic-claude-fable-5",
    "paperUrl": "",
    "dataUrl": "",
    "featured": null,
    "notes": "Evaluation conducted with 'medium' reasoning effort."
  },
  {
    "id": "anthropic-claude-fable-5-low",
    "displayName": "Claude Fable 5 (Low)",
    "modelReleaseDate": "2026-06-09T00:00:00.000Z",
    "providerId": "Anthropic",
    "modelType": "CoT",
    "modelGroup": "anthropic-claude-fable-5",
    "paperUrl": "",
    "dataUrl": "",
    "featured": null,
    "notes": "Evaluation conducted with 'low' reasoning effort."
  },
  {
    "id": "deepseek-v4-flash-0731-max",
    "displayName": "DeepSeek V4 Flash 0731 (Max)",
    "modelReleaseDate": "2026-07-31T00:00:00.000Z",
    "providerId": "DeepSeek",
    "modelType": "CoT",
    "modelGroup": "deepseek-v4-flash-0731",
    "paperUrl": "https://arxiv.org/abs/2606.19348",
    "dataUrl": "https://huggingface.co/deepseek-ai/DeepSeek-V4-Flash-0731",
    "featured": null,
    "notes": "Evaluation conducted with 'max' reasoning effort."
  },
  {
    "id": "deepseek-v4-flash-0731-high",
    "displayName": "DeepSeek V4 Flash 0731 (High)",
    "modelReleaseDate": "2026-07-31T00:00:00.000Z",
    "providerId": "DeepSeek",
    "modelType": "CoT",
    "modelGroup": "deepseek-v4-flash-0731",
    "paperUrl": "https://arxiv.org/abs/2606.19348",
    "dataUrl": "https://huggingface.co/deepseek-ai/DeepSeek-V4-Flash-0731",
    "featured": null,
    "notes": "Evaluation conducted with 'high' reasoning effort."
  },
  {
    "id": "deepseek-v4-flash-0731-low",
    "displayName": "DeepSeek V4 Flash 0731 (Low)",
    "modelReleaseDate": "2026-07-31T00:00:00.000Z",
    "providerId": "DeepSeek",
    "modelType": "CoT",
    "modelGroup": "deepseek-v4-flash-0731",
    "paperUrl": "https://arxiv.org/abs/2606.19348",
    "dataUrl": "https://huggingface.co/deepseek-ai/DeepSeek-V4-Flash-0731",
    "featured": null,
    "notes": "Evaluation conducted with 'low' reasoning effort."
  },
  {
    "id": "deepseek-v4-flash-0731-none",
    "displayName": "DeepSeek V4 Flash 0731 (None)",
    "modelReleaseDate": "2026-07-31T00:00:00.000Z",
    "providerId": "DeepSeek",
    "modelType": "CoT",
    "modelGroup": "deepseek-v4-flash-0731",
    "paperUrl": "https://arxiv.org/abs/2606.19348",
    "dataUrl": "https://huggingface.co/deepseek-ai/DeepSeek-V4-Flash-0731",
    "featured": null,
    "notes": "Evaluation conducted with 'none' reasoning effort."
  },
  {
    "id": "openai-gpt-5-6-luna-max-2026-07-30",
    "displayName": "GPT-5.6 Luna 2026-07-30 (Max)",
    "modelReleaseDate": "2026-07-30T00:00:00.000Z",
    "providerId": "OpenAI",
    "modelType": "CoT",
    "modelGroup": "openai-gpt-5-6-luna-2026-07-30",
    "paperUrl": "",
    "dataUrl": "",
    "featured": null,
    "notes": "Evaluation conducted with 'max' reasoning effort."
  },
  {
    "id": "openai-gpt-5-6-luna-xhigh-2026-07-30",
    "displayName": "GPT-5.6 Luna 2026-07-30 (XHigh)",
    "modelReleaseDate": "2026-07-30T00:00:00.000Z",
    "providerId": "OpenAI",
    "modelType": "CoT",
    "modelGroup": "openai-gpt-5-6-luna-2026-07-30",
    "paperUrl": "",
    "dataUrl": "",
    "featured": null,
    "notes": "Evaluation conducted with 'xhigh' reasoning effort."
  },
  {
    "id": "openai-gpt-5-6-luna-high-2026-07-30",
    "displayName": "GPT-5.6 Luna 2026-07-30 (High)",
    "modelReleaseDate": "2026-07-30T00:00:00.000Z",
    "providerId": "OpenAI",
    "modelType": "CoT",
    "modelGroup": "openai-gpt-5-6-luna-2026-07-30",
    "paperUrl": "",
    "dataUrl": "",
    "featured": null,
    "notes": "Evaluation conducted with 'high' reasoning effort."
  },
  {
    "id": "openai-gpt-5-6-luna-medium-2026-07-30",
    "displayName": "GPT-5.6 Luna 2026-07-30 (Medium)",
    "modelReleaseDate": "2026-07-30T00:00:00.000Z",
    "providerId": "OpenAI",
    "modelType": "CoT",
    "modelGroup": "openai-gpt-5-6-luna-2026-07-30",
    "paperUrl": "",
    "dataUrl": "",
    "featured": null,
    "notes": "Evaluation conducted with 'medium' reasoning effort."
  },
  {
    "id": "openai-gpt-5-6-luna-low-2026-07-30",
    "displayName": "GPT-5.6 Luna 2026-07-30 (Low)",
    "modelReleaseDate": "2026-07-30T00:00:00.000Z",
    "providerId": "OpenAI",
    "modelType": "CoT",
    "modelGroup": "openai-gpt-5-6-luna-2026-07-30",
    "paperUrl": "",
    "dataUrl": "",
    "featured": null,
    "notes": "Evaluation conducted with 'low' reasoning effort."
  },
  {
    "id": "openai-gpt-5-6-luna-none-2026-07-30",
    "displayName": "GPT-5.6 Luna 2026-07-30 (None)",
    "modelReleaseDate": "2026-07-30T00:00:00.000Z",
    "providerId": "OpenAI",
    "modelType": "CoT",
    "modelGroup": "openai-gpt-5-6-luna-2026-07-30",
    "paperUrl": "",
    "dataUrl": "",
    "featured": null,
    "notes": "Evaluation conducted with 'none' reasoning effort."
  },
  {
    "id": "gemini-3-5-flash-lite-high",
    "displayName": "Gemini 3.5 Flash-Lite (High)",
    "modelReleaseDate": "2026-07-21T00:00:00.000Z",
    "providerId": "Google",
    "modelType": "CoT",
    "modelGroup": "gemini-3-5-flash-lite",
    "paperUrl": "https://deepmind.google/models/gemini/flash-lite/",
    "dataUrl": "",
    "featured": null,
    "notes": "Evaluation conducted with 'high' reasoning effort."
  },
  {
    "id": "gemini-3-5-flash-lite-medium",
    "displayName": "Gemini 3.5 Flash-Lite (Medium)",
    "modelReleaseDate": "2026-07-21T00:00:00.000Z",
    "providerId": "Google",
    "modelType": "CoT",
    "modelGroup": "gemini-3-5-flash-lite",
    "paperUrl": "https://deepmind.google/models/gemini/flash-lite/",
    "dataUrl": "",
    "featured": null,
    "notes": "Evaluation conducted with 'medium' reasoning effort."
  },
  {
    "id": "gemini-3-5-flash-lite-low",
    "displayName": "Gemini 3.5 Flash-Lite (Low)",
    "modelReleaseDate": "2026-07-21T00:00:00.000Z",
    "providerId": "Google",
    "modelType": "CoT",
    "modelGroup": "gemini-3-5-flash-lite",
    "paperUrl": "https://deepmind.google/models/gemini/flash-lite/",
    "dataUrl": "",
    "featured": null,
    "notes": "Evaluation conducted with 'low' reasoning effort."
  },
  {
    "id": "gemini-3-5-flash-lite-minimal",
    "displayName": "Gemini 3.5 Flash-Lite (Minimal)",
    "modelReleaseDate": "2026-07-21T00:00:00.000Z",
    "providerId": "Google",
    "modelType": "CoT",
    "modelGroup": "gemini-3-5-flash-lite",
    "paperUrl": "https://deepmind.google/models/gemini/flash-lite/",
    "dataUrl": "",
    "featured": null,
    "notes": "Evaluation conducted with 'minimal' reasoning effort."
  },
  {
    "id": "gemini-3-6-flash-high",
    "displayName": "Gemini 3.6 Flash (High)",
    "modelReleaseDate": "2026-07-21T00:00:00.000Z",
    "providerId": "Google",
    "modelType": "CoT",
    "modelGroup": "gemini-3-6-flash",
    "paperUrl": "",
    "modelCardUrl": "https://deepmind.google/models/model-cards/gemini-3-6-flash/",
    "dataUrl": "",
    "featured": null,
    "notes": "Evaluation conducted with 'high' reasoning effort."
  },
  {
    "id": "gemini-3-6-flash-medium",
    "displayName": "Gemini 3.6 Flash (Medium)",
    "modelReleaseDate": "2026-07-21T00:00:00.000Z",
    "providerId": "Google",
    "modelType": "CoT",
    "modelGroup": "gemini-3-6-flash",
    "paperUrl": "",
    "modelCardUrl": "https://deepmind.google/models/model-cards/gemini-3-6-flash/",
    "dataUrl": "",
    "featured": null,
    "notes": "Evaluation conducted with 'medium' reasoning effort."
  },
  {
    "id": "gemini-3-6-flash-low",
    "displayName": "Gemini 3.6 Flash (Low)",
    "modelReleaseDate": "2026-07-21T00:00:00.000Z",
    "providerId": "Google",
    "modelType": "CoT",
    "modelGroup": "gemini-3-6-flash",
    "paperUrl": "",
    "modelCardUrl": "https://deepmind.google/models/model-cards/gemini-3-6-flash/",
    "dataUrl": "",
    "featured": null,
    "notes": "Evaluation conducted with 'low' reasoning effort."
  },
  {
    "id": "gemini-3-6-flash-minimal",
    "displayName": "Gemini 3.6 Flash (Minimal)",
    "modelReleaseDate": "2026-07-21T00:00:00.000Z",
    "providerId": "Google",
    "modelType": "CoT",
    "modelGroup": "gemini-3-6-flash",
    "paperUrl": "",
    "modelCardUrl": "https://deepmind.google/models/model-cards/gemini-3-6-flash/",
    "dataUrl": "",
    "featured": null,
    "notes": "Evaluation conducted with 'minimal' reasoning effort."
  },
  {
    "id": "xai-grok-4-6-xhigh",
    "displayName": "Grok 4.6 (XHigh)",
    "modelReleaseDate": "2026-08-11T00:00:00.000Z",
    "providerId": "xAI",
    "modelType": "CoT",
    "modelGroup": "xai-grok-4-6",
    "paperUrl": "",
    "dataUrl": "",
    "featured": null,
    "notes": "Evaluation conducted with 'xhigh' reasoning effort."
  },
  {
    "id": "xai-grok-4-6-high",
    "displayName": "Grok 4.6 (High)",
    "modelReleaseDate": "2026-08-11T00:00:00.000Z",
    "providerId": "xAI",
    "modelType": "CoT",
    "modelGroup": "xai-grok-4-6",
    "paperUrl": "",
    "dataUrl": "",
    "featured": null,
    "notes": "Evaluation conducted with 'high' reasoning effort."
  },
  {
    "id": "xai-grok-4-6-medium",
    "displayName": "Grok 4.6 (Medium)",
    "modelReleaseDate": "2026-08-11T00:00:00.000Z",
    "providerId": "xAI",
    "modelType": "CoT",
    "modelGroup": "xai-grok-4-6",
    "paperUrl": "",
    "dataUrl": "",
    "featured": null,
    "notes": "Evaluation conducted with 'medium' reasoning effort."
  },
  {
    "id": "xai-grok-4-6-low",
    "displayName": "Grok 4.6 (Low)",
    "modelReleaseDate": "2026-08-11T00:00:00.000Z",
    "providerId": "xAI",
    "modelType": "CoT",
    "modelGroup": "xai-grok-4-6",
    "paperUrl": "",
    "dataUrl": "",
    "featured": null,
    "notes": "Evaluation conducted with 'low' reasoning effort."
  },
  {
    "id": "google-gemini-3-7-flash-high",
    "displayName": "Gemini 3.7 Flash (High)",
    "modelReleaseDate": "2026-08-13T00:00:00.000Z",
    "providerId": "Google",
    "modelType": "CoT",
    "modelGroup": "google-gemini-3-7-flash",
    "paperUrl": "",
    "modelCardUrl": "https://deepmind.google/models/model-cards/gemini-3-7-flash/",
    "dataUrl": "",
    "featured": null,
    "notes": "Evaluation conducted with 'high' reasoning effort."
  },
  {
    "id": "google-gemini-3-7-flash-medium",
    "displayName": "Gemini 3.7 Flash (Medium)",
    "modelReleaseDate": "2026-08-13T00:00:00.000Z",
    "providerId": "Google",
    "modelType": "CoT",
    "modelGroup": "google-gemini-3-7-flash",
    "paperUrl": "",
    "modelCardUrl": "https://deepmind.google/models/model-cards/gemini-3-7-flash/",
    "dataUrl": "",
    "featured": null,
    "notes": "Evaluation conducted with 'medium' reasoning effort."
  },
  {
    "id": "google-gemini-3-7-flash-low",
    "displayName": "Gemini 3.7 Flash (Low)",
    "modelReleaseDate": "2026-08-13T00:00:00.000Z",
    "providerId": "Google",
    "modelType": "CoT",
    "modelGroup": "google-gemini-3-7-flash",
    "paperUrl": "",
    "modelCardUrl": "https://deepmind.google/models/model-cards/gemini-3-7-flash/",
    "dataUrl": "",
    "featured": null,
    "notes": "Evaluation conducted with 'low' reasoning effort."
  },
  {
    "id": "deepseek-v4-pro-0813-max",
    "displayName": "DeepSeek V4 Pro 0813 (Max)",
    "modelReleaseDate": "2026-08-13T00:00:00.000Z",
    "providerId": "DeepSeek",
    "modelType": "CoT",
    "modelGroup": "deepseek-v4-pro-0813",
    "paperUrl": "https://arxiv.org/abs/2606.19348",
    "dataUrl": "https://huggingface.co/deepseek-ai/DeepSeek-V4-Pro-0813",
    "featured": null,
    "notes": "Evaluation conducted with 'max' reasoning effort."
  },
  {
    "id": "deepseek-v4-pro-0813-high",
    "displayName": "DeepSeek V4 Pro 0813 (High)",
    "modelReleaseDate": "2026-08-13T00:00:00.000Z",
    "providerId": "DeepSeek",
    "modelType": "CoT",
    "modelGroup": "deepseek-v4-pro-0813",
    "paperUrl": "https://arxiv.org/abs/2606.19348",
    "dataUrl": "https://huggingface.co/deepseek-ai/DeepSeek-V4-Pro-0813",
    "featured": null,
    "notes": "Evaluation conducted with 'high' reasoning effort."
  },
  {
    "id": "deepseek-v4-pro-0813-low",
    "displayName": "DeepSeek V4 Pro 0813 (Low)",
    "modelReleaseDate": "2026-08-13T00:00:00.000Z",
    "providerId": "DeepSeek",
    "modelType": "CoT",
    "modelGroup": "deepseek-v4-pro-0813",
    "paperUrl": "https://arxiv.org/abs/2606.19348",
    "dataUrl": "https://huggingface.co/deepseek-ai/DeepSeek-V4-Pro-0813",
    "featured": null,
    "notes": "Evaluation conducted with 'low' reasoning effort."
  },
  {
    "id": "deepseek-v4-pro-0813-none",
    "displayName": "DeepSeek V4 Pro 0813 (None)",
    "modelReleaseDate": "2026-08-13T00:00:00.000Z",
    "providerId": "DeepSeek",
    "modelType": "CoT",
    "modelGroup": "deepseek-v4-pro-0813",
    "paperUrl": "https://arxiv.org/abs/2606.19348",
    "dataUrl": "https://huggingface.co/deepseek-ai/DeepSeek-V4-Pro-0813",
    "featured": null,
    "notes": "Evaluation conducted with 'none' reasoning effort."
  },
  {
    "id": "anthropic-claude-fable-5-1-max",
    "displayName": "Claude Fable 5.1 (Max)",
    "modelReleaseDate": "2026-09-01T00:00:00.000Z",
    "providerId": "Anthropic",
    "modelType": "CoT",
    "modelGroup": "anthropic-claude-fable-5-1",
    "paperUrl": "",
    "modelCardUrl": "",
    "dataUrl": "",
    "featured": null,
    "notes": "Evaluation conducted with 'max' reasoning effort."
  },
  {
    "id": "anthropic-claude-fable-5-1-xhigh",
    "displayName": "Claude Fable 5.1 (XHigh)",
    "modelReleaseDate": "2026-09-01T00:00:00.000Z",
    "providerId": "Anthropic",
    "modelType": "CoT",
    "modelGroup": "anthropic-claude-fable-5-1",
    "paperUrl": "",
    "modelCardUrl": "",
    "dataUrl": "",
    "featured": null,
    "notes": "Evaluation conducted with 'xhigh' reasoning effort."
  },
  {
    "id": "anthropic-claude-fable-5-1-high",
    "displayName": "Claude Fable 5.1 (High)",
    "modelReleaseDate": "2026-09-01T00:00:00.000Z",
    "providerId": "Anthropic",
    "modelType": "CoT",
    "modelGroup": "anthropic-claude-fable-5-1",
    "paperUrl": "",
    "modelCardUrl": "",
    "dataUrl": "",
    "featured": null,
    "notes": "Evaluation conducted with 'high' reasoning effort."
  },
  {
    "id": "anthropic-claude-fable-5-1-medium",
    "displayName": "Claude Fable 5.1 (Medium)",
    "modelReleaseDate": "2026-09-01T00:00:00.000Z",
    "providerId": "Anthropic",
    "modelType": "CoT",
    "modelGroup": "anthropic-claude-fable-5-1",
    "paperUrl": "",
    "modelCardUrl": "",
    "dataUrl": "",
    "featured": null,
    "notes": "Evaluation conducted with 'medium' reasoning effort."
  },
  {
    "id": "anthropic-claude-fable-5-1-low",
    "displayName": "Claude Fable 5.1 (Low)",
    "modelReleaseDate": "2026-09-01T00:00:00.000Z",
    "providerId": "Anthropic",
    "modelType": "CoT",
    "modelGroup": "anthropic-claude-fable-5-1",
    "paperUrl": "",
    "modelCardUrl": "",
    "dataUrl": "",
    "featured": null,
    "notes": "Evaluation conducted with 'low' reasoning effort."
  },
  {
    "id": "openai-gpt-6-astra-max",
    "displayName": "GPT-6 Astra (Max)",
    "modelReleaseDate": "2026-09-02T00:00:00.000Z",
    "providerId": "OpenAI",
    "modelType": "CoT",
    "modelGroup": "openai-gpt-6-astra",
    "paperUrl": "",
    "modelCardUrl": "",
    "dataUrl": "",
    "featured": null,
    "notes": "Evaluation conducted with 'max' reasoning effort."
  },
  {
    "id": "openai-gpt-6-astra-xhigh",
    "displayName": "GPT-6 Astra (XHigh)",
    "modelReleaseDate": "2026-09-02T00:00:00.000Z",
    "providerId": "OpenAI",
    "modelType": "CoT",
    "modelGroup": "openai-gpt-6-astra",
    "paperUrl": "",
    "modelCardUrl": "",
    "dataUrl": "",
    "featured": null,
    "notes": "Evaluation conducted with 'xhigh' reasoning effort."
  },
  {
    "id": "openai-gpt-6-astra-high",
    "displayName": "GPT-6 Astra (High)",
    "modelReleaseDate": "2026-09-02T00:00:00.000Z",
    "providerId": "OpenAI",
    "modelType": "CoT",
    "modelGroup": "openai-gpt-6-astra",
    "paperUrl": "",
    "modelCardUrl": "",
    "dataUrl": "",
    "featured": null,
    "notes": "Evaluation conducted with 'high' reasoning effort."
  },
  {
    "id": "openai-gpt-6-astra-medium",
    "displayName": "GPT-6 Astra (Medium)",
    "modelReleaseDate": "2026-09-02T00:00:00.000Z",
    "providerId": "OpenAI",
    "modelType": "CoT",
    "modelGroup": "openai-gpt-6-astra",
    "paperUrl": "",
    "modelCardUrl": "",
    "dataUrl": "",
    "featured": null,
    "notes": "Evaluation conducted with 'medium' reasoning effort."
  },
  {
    "id": "openai-gpt-6-astra-low",
    "displayName": "GPT-6 Astra (Low)",
    "modelReleaseDate": "2026-09-02T00:00:00.000Z",
    "providerId": "OpenAI",
    "modelType": "CoT",
    "modelGroup": "openai-gpt-6-astra",
    "paperUrl": "",
    "modelCardUrl": "",
    "dataUrl": "",
    "featured": null,
    "notes": "Evaluation conducted with 'low' reasoning effort."
  },
  {
    "id": "openai-gpt-6-astra-none",
    "displayName": "GPT-6 Astra (None)",
    "modelReleaseDate": "2026-09-02T00:00:00.000Z",
    "providerId": "OpenAI",
    "modelType": "CoT",
    "modelGroup": "openai-gpt-6-astra",
    "paperUrl": "",
    "modelCardUrl": "",
    "dataUrl": "",
    "featured": null,
    "notes": "Evaluation conducted with 'none' reasoning effort."
  },
  {
    "id": "openai-gpt-6-astra-max-provider-adapter",
    "displayName": "GPT-6 Astra - Provider Adapter (Max)",
    "modelReleaseDate": "2026-09-02T00:00:00.000Z",
    "providerId": "OpenAI",
    "modelType": "CoT",
    "modelGroup": "openai-gpt-6-astra-provider-adapter",
    "paperUrl": "",
    "modelCardUrl": "",
    "dataUrl": "",
    "featured": null,
    "notes": "Evaluation conducted with 'max' reasoning effort using the provider adapter harness."
  },
  {
    "id": "openai-gpt-6-astra-xhigh-provider-adapter",
    "displayName": "GPT-6 Astra - Provider Adapter (XHigh)",
    "modelReleaseDate": "2026-09-02T00:00:00.000Z",
    "providerId": "OpenAI",
    "modelType": "CoT",
    "modelGroup": "openai-gpt-6-astra-provider-adapter",
    "paperUrl": "",
    "modelCardUrl": "",
    "dataUrl": "",
    "featured": null,
    "notes": "Evaluation conducted with 'xhigh' reasoning effort using the provider adapter harness."
  },
  {
    "id": "openai-gpt-6-astra-high-provider-adapter",
    "displayName": "GPT-6 Astra - Provider Adapter (High)",
    "modelReleaseDate": "2026-09-02T00:00:00.000Z",
    "providerId": "OpenAI",
    "modelType": "CoT",
    "modelGroup": "openai-gpt-6-astra-provider-adapter",
    "paperUrl": "",
    "modelCardUrl": "",
    "dataUrl": "",
    "featured": null,
    "notes": "Evaluation conducted with 'high' reasoning effort using the provider adapter harness."
  },
  {
    "id": "openai-gpt-6-astra-medium-provider-adapter",
    "displayName": "GPT-6 Astra - Provider Adapter (Medium)",
    "modelReleaseDate": "2026-09-02T00:00:00.000Z",
    "providerId": "OpenAI",
    "modelType": "CoT",
    "modelGroup": "openai-gpt-6-astra-provider-adapter",
    "paperUrl": "",
    "modelCardUrl": "",
    "dataUrl": "",
    "featured": null,
    "notes": "Evaluation conducted with 'medium' reasoning effort using the provider adapter harness."
  },
  {
    "id": "openai-gpt-6-astra-low-provider-adapter",
    "displayName": "GPT-6 Astra - Provider Adapter (Low)",
    "modelReleaseDate": "2026-09-02T00:00:00.000Z",
    "providerId": "OpenAI",
    "modelType": "CoT",
    "modelGroup": "openai-gpt-6-astra-provider-adapter",
    "paperUrl": "",
    "modelCardUrl": "",
    "dataUrl": "",
    "featured": null,
    "notes": "Evaluation conducted with 'low' reasoning effort using the provider adapter harness."
  },
  {
    "id": "openai-gpt-6-astra-none-provider-adapter",
    "displayName": "GPT-6 Astra - Provider Adapter (None)",
    "modelReleaseDate": "2026-09-02T00:00:00.000Z",
    "providerId": "OpenAI",
    "modelType": "CoT",
    "modelGroup": "openai-gpt-6-astra-provider-adapter",
    "paperUrl": "",
    "modelCardUrl": "",
    "dataUrl": "",
    "featured": null,
    "notes": "Evaluation conducted with 'none' reasoning effort using the provider adapter harness."
  },
  {
    "id": "dots-studio-dots3-note-prev-max",
    "displayName": "Dots3-Note Preview (Max)",
    "modelReleaseDate": "2026-08-14T00:00:00.000Z",
    "providerId": "Dots Studio",
    "modelType": "CoT",
    "modelGroup": "dots-studio-dots3-note-prev-max",
    "paperUrl": "",
    "modelCardUrl": "",
    "dataUrl": "https://huggingface.co/dots-studio/dots3-note-prev",
    "featured": null,
    "notes": "Evaluation conducted with 'max' reasoning effort."
  },
  {
    "id": "openai-gpt-6-luna-max",
    "displayName": "GPT-6 Luna (Max)",
    "modelReleaseDate": "2026-09-22T00:00:00.000Z",
    "providerId": "OpenAI",
    "modelType": "CoT",
    "modelGroup": "openai-gpt-6-luna",
    "paperUrl": "",
    "modelCardUrl": "",
    "dataUrl": "",
    "featured": null,
    "notes": "Evaluation conducted with 'max' reasoning effort using the standard harness. Partial dataset coverage: scores use full catalog denominators; costs include recorded usage only, excluding unrecorded attempts."
  },
  {
    "id": "openai-gpt-6-luna-xhigh",
    "displayName": "GPT-6 Luna (XHigh)",
    "modelReleaseDate": "2026-09-22T00:00:00.000Z",
    "providerId": "OpenAI",
    "modelType": "CoT",
    "modelGroup": "openai-gpt-6-luna",
    "paperUrl": "",
    "modelCardUrl": "",
    "dataUrl": "",
    "featured": null,
    "notes": "Evaluation conducted with 'xhigh' reasoning effort using the standard harness."
  },
  {
    "id": "openai-gpt-6-luna-high",
    "displayName": "GPT-6 Luna (High)",
    "modelReleaseDate": "2026-09-22T00:00:00.000Z",
    "providerId": "OpenAI",
    "modelType": "CoT",
    "modelGroup": "openai-gpt-6-luna",
    "paperUrl": "",
    "modelCardUrl": "",
    "dataUrl": "",
    "featured": null,
    "notes": "Evaluation conducted with 'high' reasoning effort using the standard harness."
  },
  {
    "id": "openai-gpt-6-luna-medium",
    "displayName": "GPT-6 Luna (Medium)",
    "modelReleaseDate": "2026-09-22T00:00:00.000Z",
    "providerId": "OpenAI",
    "modelType": "CoT",
    "modelGroup": "openai-gpt-6-luna",
    "paperUrl": "",
    "modelCardUrl": "",
    "dataUrl": "",
    "featured": null,
    "notes": "Evaluation conducted with 'medium' reasoning effort using the standard harness."
  },
  {
    "id": "openai-gpt-6-luna-low",
    "displayName": "GPT-6 Luna (Low)",
    "modelReleaseDate": "2026-09-22T00:00:00.000Z",
    "providerId": "OpenAI",
    "modelType": "CoT",
    "modelGroup": "openai-gpt-6-luna",
    "paperUrl": "",
    "modelCardUrl": "",
    "dataUrl": "",
    "featured": null,
    "notes": "Evaluation conducted with 'low' reasoning effort using the standard harness."
  },
  {
    "id": "openai-gpt-6-luna-none",
    "displayName": "GPT-6 Luna (None)",
    "modelReleaseDate": "2026-09-22T00:00:00.000Z",
    "providerId": "OpenAI",
    "modelType": "CoT",
    "modelGroup": "openai-gpt-6-luna",
    "paperUrl": "",
    "modelCardUrl": "",
    "dataUrl": "",
    "featured": null,
    "notes": "Evaluation conducted with 'none' reasoning effort using the standard harness."
  },
  {
    "id": "openai-gpt-6-luna-max-provider-adapter",
    "displayName": "GPT-6 Luna - Provider Adapter (Max)",
    "modelReleaseDate": "2026-09-22T00:00:00.000Z",
    "providerId": "OpenAI",
    "modelType": "CoT",
    "modelGroup": "openai-gpt-6-luna-provider-adapter",
    "paperUrl": "",
    "modelCardUrl": "",
    "dataUrl": "",
    "featured": null,
    "notes": "Evaluation conducted with 'max' reasoning effort using the provider adapter harness. Some model actions have unmeasured token usage; reported costs are recorded-usage lower bounds, not verified full-run costs."
  },
  {
    "id": "openai-gpt-6-luna-xhigh-provider-adapter",
    "displayName": "GPT-6 Luna - Provider Adapter (XHigh)",
    "modelReleaseDate": "2026-09-22T00:00:00.000Z",
    "providerId": "OpenAI",
    "modelType": "CoT",
    "modelGroup": "openai-gpt-6-luna-provider-adapter",
    "paperUrl": "",
    "modelCardUrl": "",
    "dataUrl": "",
    "featured": null,
    "notes": "Evaluation conducted with 'xhigh' reasoning effort using the provider adapter harness. Some model actions have unmeasured token usage; reported costs are recorded-usage lower bounds, not verified full-run costs."
  },
  {
    "id": "openai-gpt-6-luna-high-provider-adapter",
    "displayName": "GPT-6 Luna - Provider Adapter (High)",
    "modelReleaseDate": "2026-09-22T00:00:00.000Z",
    "providerId": "OpenAI",
    "modelType": "CoT",
    "modelGroup": "openai-gpt-6-luna-provider-adapter",
    "paperUrl": "",
    "modelCardUrl": "",
    "dataUrl": "",
    "featured": null,
    "notes": "Evaluation conducted with 'high' reasoning effort using the provider adapter harness. Some model actions have unmeasured token usage; reported costs are recorded-usage lower bounds, not verified full-run costs."
  },
  {
    "id": "openai-gpt-6-luna-medium-provider-adapter",
    "displayName": "GPT-6 Luna - Provider Adapter (Medium)",
    "modelReleaseDate": "2026-09-22T00:00:00.000Z",
    "providerId": "OpenAI",
    "modelType": "CoT",
    "modelGroup": "openai-gpt-6-luna-provider-adapter",
    "paperUrl": "",
    "modelCardUrl": "",
    "dataUrl": "",
    "featured": null,
    "notes": "Evaluation conducted with 'medium' reasoning effort using the provider adapter harness. Some model actions have unmeasured token usage; reported costs are recorded-usage lower bounds, not verified full-run costs."
  },
  {
    "id": "openai-gpt-6-luna-low-provider-adapter",
    "displayName": "GPT-6 Luna - Provider Adapter (Low)",
    "modelReleaseDate": "2026-09-22T00:00:00.000Z",
    "providerId": "OpenAI",
    "modelType": "CoT",
    "modelGroup": "openai-gpt-6-luna-provider-adapter",
    "paperUrl": "",
    "modelCardUrl": "",
    "dataUrl": "",
    "featured": null,
    "notes": "Evaluation conducted with 'low' reasoning effort using the provider adapter harness. Some model actions have unmeasured token usage; reported costs are recorded-usage lower bounds, not verified full-run costs."
  },
  {
    "id": "openai-gpt-6-luna-none-provider-adapter",
    "displayName": "GPT-6 Luna - Provider Adapter (None)",
    "modelReleaseDate": "2026-09-22T00:00:00.000Z",
    "providerId": "OpenAI",
    "modelType": "CoT",
    "modelGroup": "openai-gpt-6-luna-provider-adapter",
    "paperUrl": "",
    "modelCardUrl": "",
    "dataUrl": "",
    "featured": null,
    "notes": "Evaluation conducted with 'none' reasoning effort using the provider adapter harness. Some model actions have unmeasured token usage; reported costs are recorded-usage lower bounds, not verified full-run costs."
  },
  {
    "id": "anthropic-claude-opus-5-5-max",
    "displayName": "Claude Opus 5.5 (Max)",
    "modelReleaseDate": "2026-09-22T00:00:00.000Z",
    "providerId": "Anthropic",
    "modelType": "CoT",
    "modelGroup": "anthropic-claude-opus-5-5",
    "paperUrl": "",
    "modelCardUrl": "",
    "dataUrl": "",
    "featured": null,
    "notes": "Evaluation conducted with 'max' reasoning effort using the standard harness. Partial dataset coverage: scores use full catalog denominators; costs include recorded usage only, excluding unrecorded attempts."
  },
  {
    "id": "anthropic-claude-opus-5-5-xhigh",
    "displayName": "Claude Opus 5.5 (XHigh)",
    "modelReleaseDate": "2026-09-22T00:00:00.000Z",
    "providerId": "Anthropic",
    "modelType": "CoT",
    "modelGroup": "anthropic-claude-opus-5-5",
    "paperUrl": "",
    "modelCardUrl": "",
    "dataUrl": "",
    "featured": null,
    "notes": "Evaluation conducted with 'xhigh' reasoning effort using the standard harness."
  },
  {
    "id": "anthropic-claude-opus-5-5-high",
    "displayName": "Claude Opus 5.5 (High)",
    "modelReleaseDate": "2026-09-22T00:00:00.000Z",
    "providerId": "Anthropic",
    "modelType": "CoT",
    "modelGroup": "anthropic-claude-opus-5-5",
    "paperUrl": "",
    "modelCardUrl": "",
    "dataUrl": "",
    "featured": null,
    "notes": "Evaluation conducted with 'high' reasoning effort using the standard harness."
  },
  {
    "id": "anthropic-claude-opus-5-5-medium",
    "displayName": "Claude Opus 5.5 (Medium)",
    "modelReleaseDate": "2026-09-22T00:00:00.000Z",
    "providerId": "Anthropic",
    "modelType": "CoT",
    "modelGroup": "anthropic-claude-opus-5-5",
    "paperUrl": "",
    "modelCardUrl": "",
    "dataUrl": "",
    "featured": null,
    "notes": "Evaluation conducted with 'medium' reasoning effort using the standard harness."
  },
  {
    "id": "anthropic-claude-opus-5-5-low",
    "displayName": "Claude Opus 5.5 (Low)",
    "modelReleaseDate": "2026-09-22T00:00:00.000Z",
    "providerId": "Anthropic",
    "modelType": "CoT",
    "modelGroup": "anthropic-claude-opus-5-5",
    "paperUrl": "",
    "modelCardUrl": "",
    "dataUrl": "",
    "featured": null,
    "notes": "Evaluation conducted with 'low' reasoning effort using the standard harness."
  },
  {
    "id": "google-gemini-3-8-flash-high",
    "displayName": "Gemini 3.8 Flash (High)",
    "modelReleaseDate": "2026-09-02T00:00:00.000Z",
    "providerId": "Google",
    "modelType": "CoT",
    "modelGroup": "google-gemini-3-8-flash",
    "paperUrl": "",
    "modelCardUrl": "https://deepmind.google/models/model-cards/gemini-3-8-flash/",
    "dataUrl": "",
    "featured": null,
    "notes": "Evaluation conducted with 'high' reasoning effort using the standard harness. Partial dataset coverage: scores use full catalog denominators; costs include recorded usage only, excluding unrecorded attempts."
  },
  {
    "id": "google-gemini-3-8-flash-medium",
    "displayName": "Gemini 3.8 Flash (Medium)",
    "modelReleaseDate": "2026-09-02T00:00:00.000Z",
    "providerId": "Google",
    "modelType": "CoT",
    "modelGroup": "google-gemini-3-8-flash",
    "paperUrl": "",
    "modelCardUrl": "https://deepmind.google/models/model-cards/gemini-3-8-flash/",
    "dataUrl": "",
    "featured": null,
    "notes": "Evaluation conducted with 'medium' reasoning effort using the standard harness."
  },
  {
    "id": "google-gemini-3-8-flash-low",
    "displayName": "Gemini 3.8 Flash (Low)",
    "modelReleaseDate": "2026-09-02T00:00:00.000Z",
    "providerId": "Google",
    "modelType": "CoT",
    "modelGroup": "google-gemini-3-8-flash",
    "paperUrl": "",
    "modelCardUrl": "https://deepmind.google/models/model-cards/gemini-3-8-flash/",
    "dataUrl": "",
    "featured": null,
    "notes": "Evaluation conducted with 'low' reasoning effort using the standard harness."
  },
  {
    "id": "google-gemini-3-8-flash-high-provider-adapter",
    "displayName": "Gemini 3.8 Flash - Provider Adapter (High)",
    "modelReleaseDate": "2026-09-02T00:00:00.000Z",
    "providerId": "Google",
    "modelType": "CoT",
    "modelGroup": "google-gemini-3-8-flash-provider-adapter",
    "paperUrl": "",
    "modelCardUrl": "https://deepmind.google/models/model-cards/gemini-3-8-flash/",
    "dataUrl": "",
    "featured": null,
    "notes": "Evaluation conducted with 'high' reasoning effort using the provider adapter harness. Partial dataset coverage: scores use full catalog denominators; costs include recorded usage only, excluding unrecorded attempts."
  },
  {
    "id": "google-gemini-3-8-flash-medium-provider-adapter",
    "displayName": "Gemini 3.8 Flash - Provider Adapter (Medium)",
    "modelReleaseDate": "2026-09-02T00:00:00.000Z",
    "providerId": "Google",
    "modelType": "CoT",
    "modelGroup": "google-gemini-3-8-flash-provider-adapter",
    "paperUrl": "",
    "modelCardUrl": "https://deepmind.google/models/model-cards/gemini-3-8-flash/",
    "dataUrl": "",
    "featured": null,
    "notes": "Evaluation conducted with 'medium' reasoning effort using the provider adapter harness."
  },
  {
    "id": "google-gemini-3-8-flash-low-provider-adapter",
    "displayName": "Gemini 3.8 Flash - Provider Adapter (Low)",
    "modelReleaseDate": "2026-09-02T00:00:00.000Z",
    "providerId": "Google",
    "modelType": "CoT",
    "modelGroup": "google-gemini-3-8-flash-provider-adapter",
    "paperUrl": "",
    "modelCardUrl": "https://deepmind.google/models/model-cards/gemini-3-8-flash/",
    "dataUrl": "",
    "featured": null,
    "notes": "Evaluation conducted with 'low' reasoning effort using the provider adapter harness. Partial dataset coverage: scores use full catalog denominators; costs include recorded usage only, excluding unrecorded attempts."
  },
  {
    "id": "openai-gpt-6-sol-max",
    "displayName": "GPT-6 Sol (Max)",
    "modelReleaseDate": "2026-09-22T00:00:00.000Z",
    "providerId": "OpenAI",
    "modelType": "CoT",
    "modelGroup": "openai-gpt-6-sol",
    "paperUrl": "",
    "modelCardUrl": "https://developers.openai.com/api/docs/models/gpt-6-sol",
    "dataUrl": "",
    "featured": null,
    "notes": "Evaluation conducted with 'max' reasoning effort using the standard harness."
  },
  {
    "id": "openai-gpt-6-sol-xhigh",
    "displayName": "GPT-6 Sol (XHigh)",
    "modelReleaseDate": "2026-09-22T00:00:00.000Z",
    "providerId": "OpenAI",
    "modelType": "CoT",
    "modelGroup": "openai-gpt-6-sol",
    "paperUrl": "",
    "modelCardUrl": "https://developers.openai.com/api/docs/models/gpt-6-sol",
    "dataUrl": "",
    "featured": null,
    "notes": "Evaluation conducted with 'xhigh' reasoning effort using the standard harness. Partial dataset coverage: scores use full catalog denominators; costs include recorded usage only, excluding unrecorded attempts."
  },
  {
    "id": "openai-gpt-6-sol-high",
    "displayName": "GPT-6 Sol (High)",
    "modelReleaseDate": "2026-09-22T00:00:00.000Z",
    "providerId": "OpenAI",
    "modelType": "CoT",
    "modelGroup": "openai-gpt-6-sol",
    "paperUrl": "",
    "modelCardUrl": "https://developers.openai.com/api/docs/models/gpt-6-sol",
    "dataUrl": "",
    "featured": null,
    "notes": "Evaluation conducted with 'high' reasoning effort using the standard harness. Partial dataset coverage: scores use full catalog denominators; costs include recorded usage only, excluding unrecorded attempts."
  },
  {
    "id": "openai-gpt-6-sol-medium",
    "displayName": "GPT-6 Sol (Medium)",
    "modelReleaseDate": "2026-09-22T00:00:00.000Z",
    "providerId": "OpenAI",
    "modelType": "CoT",
    "modelGroup": "openai-gpt-6-sol",
    "paperUrl": "",
    "modelCardUrl": "https://developers.openai.com/api/docs/models/gpt-6-sol",
    "dataUrl": "",
    "featured": null,
    "notes": "Evaluation conducted with 'medium' reasoning effort using the standard harness. Partial dataset coverage: scores use full catalog denominators; costs include recorded usage only, excluding unrecorded attempts."
  },
  {
    "id": "openai-gpt-6-sol-low",
    "displayName": "GPT-6 Sol (Low)",
    "modelReleaseDate": "2026-09-22T00:00:00.000Z",
    "providerId": "OpenAI",
    "modelType": "CoT",
    "modelGroup": "openai-gpt-6-sol",
    "paperUrl": "",
    "modelCardUrl": "https://developers.openai.com/api/docs/models/gpt-6-sol",
    "dataUrl": "",
    "featured": null,
    "notes": "Evaluation conducted with 'low' reasoning effort using the standard harness. Partial dataset coverage: scores use full catalog denominators; costs include recorded usage only, excluding unrecorded attempts."
  },
  {
    "id": "openai-gpt-6-sol-none",
    "displayName": "GPT-6 Sol (None)",
    "modelReleaseDate": "2026-09-22T00:00:00.000Z",
    "providerId": "OpenAI",
    "modelType": "CoT",
    "modelGroup": "openai-gpt-6-sol",
    "paperUrl": "",
    "modelCardUrl": "https://developers.openai.com/api/docs/models/gpt-6-sol",
    "dataUrl": "",
    "featured": null,
    "notes": "Evaluation conducted with 'none' reasoning effort using the standard harness. Partial dataset coverage: scores use full catalog denominators; costs include recorded usage only, excluding unrecorded attempts."
  },
  {
    "id": "openai-gpt-6-sol-max-provider-adapter",
    "displayName": "GPT-6 Sol - Provider Adapter (Max)",
    "modelReleaseDate": "2026-09-22T00:00:00.000Z",
    "providerId": "OpenAI",
    "modelType": "CoT",
    "modelGroup": "openai-gpt-6-sol-provider-adapter",
    "paperUrl": "",
    "modelCardUrl": "https://developers.openai.com/api/docs/models/gpt-6-sol",
    "dataUrl": "",
    "featured": null,
    "notes": "Evaluation conducted with 'max' reasoning effort using the provider adapter harness. Partial dataset coverage: scores use full catalog denominators; costs include recorded usage only, excluding unrecorded attempts. Some model actions have unmeasured token usage; reported costs are recorded-usage lower bounds, not verified full-run costs."
  },
  {
    "id": "openai-gpt-6-sol-xhigh-provider-adapter",
    "displayName": "GPT-6 Sol - Provider Adapter (XHigh)",
    "modelReleaseDate": "2026-09-22T00:00:00.000Z",
    "providerId": "OpenAI",
    "modelType": "CoT",
    "modelGroup": "openai-gpt-6-sol-provider-adapter",
    "paperUrl": "",
    "modelCardUrl": "https://developers.openai.com/api/docs/models/gpt-6-sol",
    "dataUrl": "",
    "featured": null,
    "notes": "Evaluation conducted with 'xhigh' reasoning effort using the provider adapter harness. Some model actions have unmeasured token usage; reported costs are recorded-usage lower bounds, not verified full-run costs."
  },
  {
    "id": "openai-gpt-6-sol-high-provider-adapter",
    "displayName": "GPT-6 Sol - Provider Adapter (High)",
    "modelReleaseDate": "2026-09-22T00:00:00.000Z",
    "providerId": "OpenAI",
    "modelType": "CoT",
    "modelGroup": "openai-gpt-6-sol-provider-adapter",
    "paperUrl": "",
    "modelCardUrl": "https://developers.openai.com/api/docs/models/gpt-6-sol",
    "dataUrl": "",
    "featured": null,
    "notes": "Evaluation conducted with 'high' reasoning effort using the provider adapter harness. Some model actions have unmeasured token usage; reported costs are recorded-usage lower bounds, not verified full-run costs."
  },
  {
    "id": "openai-gpt-6-sol-medium-provider-adapter",
    "displayName": "GPT-6 Sol - Provider Adapter (Medium)",
    "modelReleaseDate": "2026-09-22T00:00:00.000Z",
    "providerId": "OpenAI",
    "modelType": "CoT",
    "modelGroup": "openai-gpt-6-sol-provider-adapter",
    "paperUrl": "",
    "modelCardUrl": "https://developers.openai.com/api/docs/models/gpt-6-sol",
    "dataUrl": "",
    "featured": null,
    "notes": "Evaluation conducted with 'medium' reasoning effort using the provider adapter harness. Some model actions have unmeasured token usage; reported costs are recorded-usage lower bounds, not verified full-run costs."
  },
  {
    "id": "openai-gpt-6-sol-low-provider-adapter",
    "displayName": "GPT-6 Sol - Provider Adapter (Low)",
    "modelReleaseDate": "2026-09-22T00:00:00.000Z",
    "providerId": "OpenAI",
    "modelType": "CoT",
    "modelGroup": "openai-gpt-6-sol-provider-adapter",
    "paperUrl": "",
    "modelCardUrl": "https://developers.openai.com/api/docs/models/gpt-6-sol",
    "dataUrl": "",
    "featured": null,
    "notes": "Evaluation conducted with 'low' reasoning effort using the provider adapter harness."
  },
  {
    "id": "openai-gpt-6-sol-none-provider-adapter",
    "displayName": "GPT-6 Sol - Provider Adapter (None)",
    "modelReleaseDate": "2026-09-22T00:00:00.000Z",
    "providerId": "OpenAI",
    "modelType": "CoT",
    "modelGroup": "openai-gpt-6-sol-provider-adapter",
    "paperUrl": "",
    "modelCardUrl": "https://developers.openai.com/api/docs/models/gpt-6-sol",
    "dataUrl": "",
    "featured": null,
    "notes": "Evaluation conducted with 'none' reasoning effort using the provider adapter harness. Partial dataset coverage: scores use full catalog denominators; costs include recorded usage only, excluding unrecorded attempts. Some model actions have unmeasured token usage; reported costs are recorded-usage lower bounds, not verified full-run costs."
  },
  {
    "id": "zai-glm-5-3-flash-max",
    "displayName": "GLM-5.3-Flash (Max)",
    "modelReleaseDate": "2026-08-26T00:00:00.000Z",
    "providerId": "Z.ai",
    "modelType": "CoT",
    "modelGroup": "zai-glm-5-3-flash",
    "paperUrl": "",
    "modelCardUrl": "",
    "dataUrl": "https://huggingface.co/zai-org/GLM-5.3-Flash",
    "featured": null,
    "notes": "Evaluation conducted with 'max' reasoning effort."
  },
  {
    "id": "zai-glm-5-3-flash-high",
    "displayName": "GLM-5.3-Flash (High)",
    "modelReleaseDate": "2026-08-26T00:00:00.000Z",
    "providerId": "Z.ai",
    "modelType": "CoT",
    "modelGroup": "zai-glm-5-3-flash",
    "paperUrl": "",
    "modelCardUrl": "",
    "dataUrl": "https://huggingface.co/zai-org/GLM-5.3-Flash",
    "featured": null,
    "notes": "Evaluation conducted with 'high' reasoning effort."
  },
  {
    "id": "zai-glm-5-3-flash-low",
    "displayName": "GLM-5.3-Flash (Low)",
    "modelReleaseDate": "2026-08-26T00:00:00.000Z",
    "providerId": "Z.ai",
    "modelType": "CoT",
    "modelGroup": "zai-glm-5-3-flash",
    "paperUrl": "",
    "modelCardUrl": "",
    "dataUrl": "https://huggingface.co/zai-org/GLM-5.3-Flash",
    "featured": null,
    "notes": "Evaluation conducted with 'low' reasoning effort."
  },
  {
    "id": "openai-gpt-6-1-sol-max",
    "displayName": "GPT-6.1 Sol (Max)",
    "modelReleaseDate": "2026-09-29T00:00:00.000Z",
    "providerId": "OpenAI",
    "modelType": "CoT",
    "modelGroup": "openai-gpt-6-1-sol",
    "paperUrl": "",
    "modelCardUrl": "https://developers.openai.com/api/docs/models/gpt-6.1-sol",
    "dataUrl": "",
    "featured": null,
    "notes": "Evaluation conducted with 'max' reasoning effort using the standard harness."
  },
  {
    "id": "openai-gpt-6-1-sol-xhigh",
    "displayName": "GPT-6.1 Sol (XHigh)",
    "modelReleaseDate": "2026-09-29T00:00:00.000Z",
    "providerId": "OpenAI",
    "modelType": "CoT",
    "modelGroup": "openai-gpt-6-1-sol",
    "paperUrl": "",
    "modelCardUrl": "https://developers.openai.com/api/docs/models/gpt-6.1-sol",
    "dataUrl": "",
    "featured": null,
    "notes": "Evaluation conducted with 'xhigh' reasoning effort using the standard harness."
  },
  {
    "id": "openai-gpt-6-1-sol-high",
    "displayName": "GPT-6.1 Sol (High)",
    "modelReleaseDate": "2026-09-29T00:00:00.000Z",
    "providerId": "OpenAI",
    "modelType": "CoT",
    "modelGroup": "openai-gpt-6-1-sol",
    "paperUrl": "",
    "modelCardUrl": "https://developers.openai.com/api/docs/models/gpt-6.1-sol",
    "dataUrl": "",
    "featured": null,
    "notes": "Evaluation conducted with 'high' reasoning effort using the standard harness."
  },
  {
    "id": "openai-gpt-6-1-sol-medium",
    "displayName": "GPT-6.1 Sol (Medium)",
    "modelReleaseDate": "2026-09-29T00:00:00.000Z",
    "providerId": "OpenAI",
    "modelType": "CoT",
    "modelGroup": "openai-gpt-6-1-sol",
    "paperUrl": "",
    "modelCardUrl": "https://developers.openai.com/api/docs/models/gpt-6.1-sol",
    "dataUrl": "",
    "featured": null,
    "notes": "Evaluation conducted with 'medium' reasoning effort using the standard harness."
  },
  {
    "id": "openai-gpt-6-1-sol-low",
    "displayName": "GPT-6.1 Sol (Low)",
    "modelReleaseDate": "2026-09-29T00:00:00.000Z",
    "providerId": "OpenAI",
    "modelType": "CoT",
    "modelGroup": "openai-gpt-6-1-sol",
    "paperUrl": "",
    "modelCardUrl": "https://developers.openai.com/api/docs/models/gpt-6.1-sol",
    "dataUrl": "",
    "featured": null,
    "notes": "Evaluation conducted with 'low' reasoning effort using the standard harness."
  },
  {
    "id": "openai-gpt-6-1-sol-max-provider-adapter",
    "displayName": "GPT-6.1 Sol - Provider Adapter (Max)",
    "modelReleaseDate": "2026-09-29T00:00:00.000Z",
    "providerId": "OpenAI",
    "modelType": "CoT",
    "modelGroup": "openai-gpt-6-1-sol-provider-adapter",
    "paperUrl": "",
    "modelCardUrl": "https://developers.openai.com/api/docs/models/gpt-6.1-sol",
    "dataUrl": "",
    "featured": null,
    "notes": "Evaluation conducted with 'max' reasoning effort using the provider adapter harness. Some model actions have unmeasured token usage; reported costs are recorded-usage lower bounds, not verified full-run costs."
  },
  {
    "id": "openai-gpt-6-1-sol-xhigh-provider-adapter",
    "displayName": "GPT-6.1 Sol - Provider Adapter (XHigh)",
    "modelReleaseDate": "2026-09-29T00:00:00.000Z",
    "providerId": "OpenAI",
    "modelType": "CoT",
    "modelGroup": "openai-gpt-6-1-sol-provider-adapter",
    "paperUrl": "",
    "modelCardUrl": "https://developers.openai.com/api/docs/models/gpt-6.1-sol",
    "dataUrl": "",
    "featured": null,
    "notes": "Evaluation conducted with 'xhigh' reasoning effort using the provider adapter harness. Some model actions have unmeasured token usage; reported costs are recorded-usage lower bounds, not verified full-run costs."
  },
  {
    "id": "openai-gpt-6-1-sol-high-provider-adapter",
    "displayName": "GPT-6.1 Sol - Provider Adapter (High)",
    "modelReleaseDate": "2026-09-29T00:00:00.000Z",
    "providerId": "OpenAI",
    "modelType": "CoT",
    "modelGroup": "openai-gpt-6-1-sol-provider-adapter",
    "paperUrl": "",
    "modelCardUrl": "https://developers.openai.com/api/docs/models/gpt-6.1-sol",
    "dataUrl": "",
    "featured": null,
    "notes": "Evaluation conducted with 'high' reasoning effort using the provider adapter harness. Partial dataset coverage: scores use full catalog denominators; costs include recorded usage only, excluding unrecorded attempts. Some model actions have unmeasured token usage; reported costs are recorded-usage lower bounds, not verified full-run costs."
  },
  {
    "id": "openai-gpt-6-1-sol-medium-provider-adapter",
    "displayName": "GPT-6.1 Sol - Provider Adapter (Medium)",
    "modelReleaseDate": "2026-09-29T00:00:00.000Z",
    "providerId": "OpenAI",
    "modelType": "CoT",
    "modelGroup": "openai-gpt-6-1-sol-provider-adapter",
    "paperUrl": "",
    "modelCardUrl": "https://developers.openai.com/api/docs/models/gpt-6.1-sol",
    "dataUrl": "",
    "featured": null,
    "notes": "Evaluation conducted with 'medium' reasoning effort using the provider adapter harness. Partial dataset coverage: scores use full catalog denominators; costs include recorded usage only, excluding unrecorded attempts. Some model actions have unmeasured token usage; reported costs are recorded-usage lower bounds, not verified full-run costs."
  },
  {
    "id": "openai-gpt-6-1-sol-low-provider-adapter",
    "displayName": "GPT-6.1 Sol - Provider Adapter (Low)",
    "modelReleaseDate": "2026-09-29T00:00:00.000Z",
    "providerId": "OpenAI",
    "modelType": "CoT",
    "modelGroup": "openai-gpt-6-1-sol-provider-adapter",
    "paperUrl": "",
    "modelCardUrl": "https://developers.openai.com/api/docs/models/gpt-6.1-sol",
    "dataUrl": "",
    "featured": null,
    "notes": "Evaluation conducted with 'low' reasoning effort using the provider adapter harness. Partial dataset coverage: scores use full catalog denominators; costs include recorded usage only, excluding unrecorded attempts. Some model actions have unmeasured token usage; reported costs are recorded-usage lower bounds, not verified full-run costs."
  },
  {
    "id": "alibaba-qwen3-8-27b-xhigh",
    "displayName": "Qwen3.8-27B (XHigh)",
    "modelReleaseDate": "2026-08-14T00:00:00.000Z",
    "providerId": "Alibaba",
    "modelType": "CoT",
    "modelGroup": "alibaba-qwen3-8-27b",
    "paperUrl": "",
    "modelCardUrl": "",
    "dataUrl": "https://huggingface.co/Qwen/Qwen3.8-27B",
    "featured": null,
    "notes": "Evaluation conducted with 'xhigh' reasoning effort."
  },
  {
    "id": "alibaba-qwen3-8-27b-medium",
    "displayName": "Qwen3.8-27B (Medium)",
    "modelReleaseDate": "2026-08-14T00:00:00.000Z",
    "providerId": "Alibaba",
    "modelType": "CoT",
    "modelGroup": "alibaba-qwen3-8-27b",
    "paperUrl": "",
    "modelCardUrl": "",
    "dataUrl": "https://huggingface.co/Qwen/Qwen3.8-27B",
    "featured": null,
    "notes": "Evaluation conducted with 'medium' reasoning effort."
  },
  {
    "id": "alibaba-qwen3-8-27b-low",
    "displayName": "Qwen3.8-27B (Low)",
    "modelReleaseDate": "2026-08-14T00:00:00.000Z",
    "providerId": "Alibaba",
    "modelType": "CoT",
    "modelGroup": "alibaba-qwen3-8-27b",
    "paperUrl": "",
    "modelCardUrl": "",
    "dataUrl": "https://huggingface.co/Qwen/Qwen3.8-27B",
    "featured": null,
    "notes": "Evaluation conducted with 'low' reasoning effort."
  },
  {
    "id": "alibaba-qwen3-8-27b-none",
    "displayName": "Qwen3.8-27B (None)",
    "modelReleaseDate": "2026-08-14T00:00:00.000Z",
    "providerId": "Alibaba",
    "modelType": "CoT",
    "modelGroup": "alibaba-qwen3-8-27b",
    "paperUrl": "",
    "modelCardUrl": "",
    "dataUrl": "https://huggingface.co/Qwen/Qwen3.8-27B",
    "featured": null,
    "notes": "Evaluation conducted with 'none' reasoning effort."
  },
  {
    "id": "deepseek-v4-1-flash-max",
    "displayName": "DeepSeek V4.1 Flash (Max)",
    "modelReleaseDate": "2026-09-10T00:00:00.000Z",
    "providerId": "DeepSeek",
    "modelType": "CoT",
    "modelGroup": "deepseek-v4-1-flash",
    "paperUrl": "",
    "modelCardUrl": "",
    "dataUrl": "https://huggingface.co/deepseek-ai/DeepSeek-V4.1-Flash",
    "featured": null,
    "notes": "Evaluation conducted with 'max' reasoning effort using the standard harness. Partial dataset coverage: scores use full catalog denominators; costs include recorded usage only, excluding unrecorded attempts."
  },
  {
    "id": "deepseek-v4-1-flash-high",
    "displayName": "DeepSeek V4.1 Flash (High)",
    "modelReleaseDate": "2026-09-10T00:00:00.000Z",
    "providerId": "DeepSeek",
    "modelType": "CoT",
    "modelGroup": "deepseek-v4-1-flash",
    "paperUrl": "",
    "modelCardUrl": "",
    "dataUrl": "https://huggingface.co/deepseek-ai/DeepSeek-V4.1-Flash",
    "featured": null,
    "notes": "Evaluation conducted with 'high' reasoning effort using the standard harness. Partial dataset coverage: scores use full catalog denominators; costs include recorded usage only, excluding unrecorded attempts."
  },
  {
    "id": "deepseek-v4-1-flash-low",
    "displayName": "DeepSeek V4.1 Flash (Low)",
    "modelReleaseDate": "2026-09-10T00:00:00.000Z",
    "providerId": "DeepSeek",
    "modelType": "CoT",
    "modelGroup": "deepseek-v4-1-flash",
    "paperUrl": "",
    "modelCardUrl": "",
    "dataUrl": "https://huggingface.co/deepseek-ai/DeepSeek-V4.1-Flash",
    "featured": null,
    "notes": "Evaluation conducted with 'low' reasoning effort using the standard harness."
  },
  {
    "id": "deepseek-v4-1-flash-none",
    "displayName": "DeepSeek V4.1 Flash (None)",
    "modelReleaseDate": "2026-09-10T00:00:00.000Z",
    "providerId": "DeepSeek",
    "modelType": "CoT",
    "modelGroup": "deepseek-v4-1-flash",
    "paperUrl": "",
    "modelCardUrl": "",
    "dataUrl": "https://huggingface.co/deepseek-ai/DeepSeek-V4.1-Flash",
    "featured": null,
    "notes": "Evaluation conducted with 'none' reasoning effort using the standard harness."
  },
  {
    "id": "xai-grok-4-7-xhigh",
    "displayName": "Grok 4.7 (XHigh)",
    "modelReleaseDate": "2026-09-21T00:00:00.000Z",
    "providerId": "xAI",
    "modelType": "CoT",
    "modelGroup": "xai-grok-4-7",
    "paperUrl": "",
    "modelCardUrl": "",
    "dataUrl": "",
    "featured": null,
    "notes": "Evaluation conducted with 'xhigh' reasoning effort using the standard harness. Partial dataset coverage: scores use full catalog denominators; costs include recorded usage only, excluding unrecorded attempts."
  },
  {
    "id": "xai-grok-4-7-high",
    "displayName": "Grok 4.7 (High)",
    "modelReleaseDate": "2026-09-21T00:00:00.000Z",
    "providerId": "xAI",
    "modelType": "CoT",
    "modelGroup": "xai-grok-4-7",
    "paperUrl": "",
    "modelCardUrl": "",
    "dataUrl": "",
    "featured": null,
    "notes": "Evaluation conducted with 'high' reasoning effort using the standard harness. Partial dataset coverage: scores use full catalog denominators; costs include recorded usage only, excluding unrecorded attempts."
  },
  {
    "id": "xai-grok-4-7-medium",
    "displayName": "Grok 4.7 (Medium)",
    "modelReleaseDate": "2026-09-21T00:00:00.000Z",
    "providerId": "xAI",
    "modelType": "CoT",
    "modelGroup": "xai-grok-4-7",
    "paperUrl": "",
    "modelCardUrl": "",
    "dataUrl": "",
    "featured": null,
    "notes": "Evaluation conducted with 'medium' reasoning effort using the standard harness. Partial dataset coverage: scores use full catalog denominators; costs include recorded usage only, excluding unrecorded attempts."
  },
  {
    "id": "xai-grok-4-7-low",
    "displayName": "Grok 4.7 (Low)",
    "modelReleaseDate": "2026-09-21T00:00:00.000Z",
    "providerId": "xAI",
    "modelType": "CoT",
    "modelGroup": "xai-grok-4-7",
    "paperUrl": "",
    "modelCardUrl": "",
    "dataUrl": "",
    "featured": null,
    "notes": "Evaluation conducted with 'low' reasoning effort using the standard harness."
  },
  {
    "id": "xai-grok-4-7-xhigh-provider-adapter",
    "displayName": "Grok 4.7 - Provider Adapter (XHigh)",
    "modelReleaseDate": "2026-09-21T00:00:00.000Z",
    "providerId": "xAI",
    "modelType": "CoT",
    "modelGroup": "xai-grok-4-7-provider-adapter",
    "paperUrl": "",
    "modelCardUrl": "",
    "dataUrl": "",
    "featured": null,
    "notes": "Evaluation conducted with 'xhigh' reasoning effort using the provider adapter harness. Partial dataset coverage: scores use full catalog denominators; costs include recorded usage only, excluding unrecorded attempts."
  },
  {
    "id": "xai-grok-4-7-high-provider-adapter",
    "displayName": "Grok 4.7 - Provider Adapter (High)",
    "modelReleaseDate": "2026-09-21T00:00:00.000Z",
    "providerId": "xAI",
    "modelType": "CoT",
    "modelGroup": "xai-grok-4-7-provider-adapter",
    "paperUrl": "",
    "modelCardUrl": "",
    "dataUrl": "",
    "featured": null,
    "notes": "Evaluation conducted with 'high' reasoning effort using the provider adapter harness. Partial dataset coverage: scores use full catalog denominators; costs include recorded usage only, excluding unrecorded attempts."
  },
  {
    "id": "xai-grok-4-7-medium-provider-adapter",
    "displayName": "Grok 4.7 - Provider Adapter (Medium)",
    "modelReleaseDate": "2026-09-21T00:00:00.000Z",
    "providerId": "xAI",
    "modelType": "CoT",
    "modelGroup": "xai-grok-4-7-provider-adapter",
    "paperUrl": "",
    "modelCardUrl": "",
    "dataUrl": "",
    "featured": null,
    "notes": "Evaluation conducted with 'medium' reasoning effort using the provider adapter harness."
  },
  {
    "id": "xai-grok-4-7-low-provider-adapter",
    "displayName": "Grok 4.7 - Provider Adapter (Low)",
    "modelReleaseDate": "2026-09-21T00:00:00.000Z",
    "providerId": "xAI",
    "modelType": "CoT",
    "modelGroup": "xai-grok-4-7-provider-adapter",
    "paperUrl": "",
    "modelCardUrl": "",
    "dataUrl": "",
    "featured": null,
    "notes": "Evaluation conducted with 'low' reasoning effort using the provider adapter harness."
  },
  {
    "id": "deepseek-v4-1-flash-max-provider-adapter",
    "displayName": "DeepSeek V4.1 Flash - Provider Adapter (Max)",
    "modelReleaseDate": "2026-09-10T00:00:00.000Z",
    "providerId": "DeepSeek",
    "modelType": "CoT",
    "modelGroup": "deepseek-v4-1-flash-provider-adapter",
    "paperUrl": "",
    "modelCardUrl": "",
    "dataUrl": "https://huggingface.co/deepseek-ai/DeepSeek-V4.1-Flash",
    "featured": null,
    "notes": "Evaluation conducted with 'max' reasoning effort using the provider adapter harness."
  },
  {
    "id": "deepseek-v4-1-flash-high-provider-adapter",
    "displayName": "DeepSeek V4.1 Flash - Provider Adapter (High)",
    "modelReleaseDate": "2026-09-10T00:00:00.000Z",
    "providerId": "DeepSeek",
    "modelType": "CoT",
    "modelGroup": "deepseek-v4-1-flash-provider-adapter",
    "paperUrl": "",
    "modelCardUrl": "",
    "dataUrl": "https://huggingface.co/deepseek-ai/DeepSeek-V4.1-Flash",
    "featured": null,
    "notes": "Evaluation conducted with 'high' reasoning effort using the provider adapter harness."
  },
  {
    "id": "deepseek-v4-1-flash-low-provider-adapter",
    "displayName": "DeepSeek V4.1 Flash - Provider Adapter (Low)",
    "modelReleaseDate": "2026-09-10T00:00:00.000Z",
    "providerId": "DeepSeek",
    "modelType": "CoT",
    "modelGroup": "deepseek-v4-1-flash-provider-adapter",
    "paperUrl": "",
    "modelCardUrl": "",
    "dataUrl": "https://huggingface.co/deepseek-ai/DeepSeek-V4.1-Flash",
    "featured": null,
    "notes": "Evaluation conducted with 'low' reasoning effort using the provider adapter harness. Some model actions have unmeasured token usage; reported costs are recorded-usage lower bounds, not verified full-run costs."
  },
  {
    "id": "deepseek-v4-1-flash-none-provider-adapter",
    "displayName": "DeepSeek V4.1 Flash - Provider Adapter (None)",
    "modelReleaseDate": "2026-09-10T00:00:00.000Z",
    "providerId": "DeepSeek",
    "modelType": "CoT",
    "modelGroup": "deepseek-v4-1-flash-provider-adapter",
    "paperUrl": "",
    "modelCardUrl": "",
    "dataUrl": "https://huggingface.co/deepseek-ai/DeepSeek-V4.1-Flash",
    "featured": null,
    "notes": "Evaluation conducted with 'none' reasoning effort using the provider adapter harness."
  },
  {
    "id": "anthropic-claude-haiku-5-5-max",
    "displayName": "Claude Haiku 5.5 (Max)",
    "modelReleaseDate": "2026-10-07T00:00:00.000Z",
    "providerId": "Anthropic",
    "modelType": "CoT",
    "modelGroup": "anthropic-claude-haiku-5-5",
    "paperUrl": "",
    "modelCardUrl": "",
    "dataUrl": "",
    "featured": null,
    "notes": "Evaluation conducted with 'max' reasoning effort using the standard harness. Partial dataset coverage: scores use full catalog denominators; costs include recorded usage only, excluding unrecorded attempts."
  },
  {
    "id": "anthropic-claude-haiku-5-5-xhigh",
    "displayName": "Claude Haiku 5.5 (XHigh)",
    "modelReleaseDate": "2026-10-07T00:00:00.000Z",
    "providerId": "Anthropic",
    "modelType": "CoT",
    "modelGroup": "anthropic-claude-haiku-5-5",
    "paperUrl": "",
    "modelCardUrl": "",
    "dataUrl": "",
    "featured": null,
    "notes": "Evaluation conducted with 'xhigh' reasoning effort using the standard harness. Partial dataset coverage: scores use full catalog denominators; costs include recorded usage only, excluding unrecorded attempts."
  },
  {
    "id": "anthropic-claude-haiku-5-5-high",
    "displayName": "Claude Haiku 5.5 (High)",
    "modelReleaseDate": "2026-10-07T00:00:00.000Z",
    "providerId": "Anthropic",
    "modelType": "CoT",
    "modelGroup": "anthropic-claude-haiku-5-5",
    "paperUrl": "",
    "modelCardUrl": "",
    "dataUrl": "",
    "featured": null,
    "notes": "Evaluation conducted with 'high' reasoning effort using the standard harness. Partial dataset coverage: scores use full catalog denominators; costs include recorded usage only, excluding unrecorded attempts."
  },
  {
    "id": "anthropic-claude-haiku-5-5-medium",
    "displayName": "Claude Haiku 5.5 (Medium)",
    "modelReleaseDate": "2026-10-07T00:00:00.000Z",
    "providerId": "Anthropic",
    "modelType": "CoT",
    "modelGroup": "anthropic-claude-haiku-5-5",
    "paperUrl": "",
    "modelCardUrl": "",
    "dataUrl": "",
    "featured": null,
    "notes": "Evaluation conducted with 'medium' reasoning effort using the standard harness. Partial dataset coverage: scores use full catalog denominators; costs include recorded usage only, excluding unrecorded attempts."
  },
  {
    "id": "anthropic-claude-haiku-5-5-low",
    "displayName": "Claude Haiku 5.5 (Low)",
    "modelReleaseDate": "2026-10-07T00:00:00.000Z",
    "providerId": "Anthropic",
    "modelType": "CoT",
    "modelGroup": "anthropic-claude-haiku-5-5",
    "paperUrl": "",
    "modelCardUrl": "",
    "dataUrl": "",
    "featured": null,
    "notes": "Evaluation conducted with 'low' reasoning effort using the standard harness. Partial dataset coverage: scores use full catalog denominators; costs include recorded usage only, excluding unrecorded attempts."
  },
  {
    "id": "anthropic-claude-haiku-5-5-max-provider-adapter",
    "displayName": "Claude Haiku 5.5 - Provider Adapter (Max)",
    "modelReleaseDate": "2026-10-07T00:00:00.000Z",
    "providerId": "Anthropic",
    "modelType": "CoT",
    "modelGroup": "anthropic-claude-haiku-5-5-provider-adapter",
    "paperUrl": "",
    "modelCardUrl": "",
    "dataUrl": "",
    "featured": null,
    "notes": "Evaluation conducted with 'max' reasoning effort using the provider adapter harness. Partial dataset coverage: scores use full catalog denominators; costs include recorded usage only, excluding unrecorded attempts."
  },
  {
    "id": "google-gemini-4-argon-high",
    "displayName": "Gemini 4 Argon (High)",
    "modelReleaseDate": "2026-10-13T00:00:00.000Z",
    "providerId": "Google",
    "modelType": "CoT",
    "modelGroup": "google-gemini-4-argon",
    "paperUrl": "",
    "modelCardUrl": "",
    "dataUrl": "",
    "featured": null,
    "notes": "Evaluation conducted with 'high' reasoning effort using the standard harness. Partial dataset coverage: scores use full catalog denominators; costs include recorded usage only, excluding unrecorded attempts."
  },
  {
    "id": "google-gemini-4-argon-medium",
    "displayName": "Gemini 4 Argon (Medium)",
    "modelReleaseDate": "2026-10-13T00:00:00.000Z",
    "providerId": "Google",
    "modelType": "CoT",
    "modelGroup": "google-gemini-4-argon",
    "paperUrl": "",
    "modelCardUrl": "",
    "dataUrl": "",
    "featured": null,
    "notes": "Evaluation conducted with 'medium' reasoning effort using the standard harness. Partial dataset coverage: scores use full catalog denominators; costs include recorded usage only, excluding unrecorded attempts."
  },
  {
    "id": "google-gemini-4-argon-low",
    "displayName": "Gemini 4 Argon (Low)",
    "modelReleaseDate": "2026-10-13T00:00:00.000Z",
    "providerId": "Google",
    "modelType": "CoT",
    "modelGroup": "google-gemini-4-argon",
    "paperUrl": "",
    "modelCardUrl": "",
    "dataUrl": "",
    "featured": null,
    "notes": "Evaluation conducted with 'low' reasoning effort using the standard harness. Partial dataset coverage: scores use full catalog denominators; costs include recorded usage only, excluding unrecorded attempts."
  },
  {
    "id": "google-gemini-4-argon-high-provider-adapter",
    "displayName": "Gemini 4 Argon - Provider Adapter (High)",
    "modelReleaseDate": "2026-10-13T00:00:00.000Z",
    "providerId": "Google",
    "modelType": "CoT",
    "modelGroup": "google-gemini-4-argon-provider-adapter",
    "paperUrl": "",
    "modelCardUrl": "",
    "dataUrl": "",
    "featured": null,
    "notes": "Evaluation conducted with 'high' reasoning effort using the provider adapter harness. Partial dataset coverage: scores use full catalog denominators; costs include recorded usage only, excluding unrecorded attempts."
  },
  {
    "id": "google-gemini-4-argon-medium-provider-adapter",
    "displayName": "Gemini 4 Argon - Provider Adapter (Medium)",
    "modelReleaseDate": "2026-10-13T00:00:00.000Z",
    "providerId": "Google",
    "modelType": "CoT",
    "modelGroup": "google-gemini-4-argon-provider-adapter",
    "paperUrl": "",
    "modelCardUrl": "",
    "dataUrl": "",
    "featured": null,
    "notes": "Evaluation conducted with 'medium' reasoning effort using the provider adapter harness. Partial dataset coverage: scores use full catalog denominators; costs include recorded usage only, excluding unrecorded attempts."
  },
  {
    "id": "google-gemini-4-argon-low-provider-adapter",
    "displayName": "Gemini 4 Argon - Provider Adapter (Low)",
    "modelReleaseDate": "2026-10-13T00:00:00.000Z",
    "providerId": "Google",
    "modelType": "CoT",
    "modelGroup": "google-gemini-4-argon-provider-adapter",
    "paperUrl": "",
    "modelCardUrl": "",
    "dataUrl": "",
    "featured": null,
    "notes": "Evaluation conducted with 'low' reasoning effort using the provider adapter harness. Partial dataset coverage: scores use full catalog denominators; costs include recorded usage only, excluding unrecorded attempts."
  },
  {
    "id": "reflection-beam-1-0-xhigh",
    "displayName": "Beam 1.0 (XHigh)",
    "modelReleaseDate": "2026-10-16T00:00:00.000Z",
    "providerId": "Reflection",
    "modelType": "CoT",
    "modelGroup": "reflection-beam-1-0",
    "paperUrl": "",
    "modelCardUrl": "",
    "dataUrl": "https://huggingface.co/reflection/Beam-1.0-501B-A23B",
    "featured": null,
    "notes": "Evaluation conducted with 'xhigh' reasoning effort using the standard harness."
  },
  {
    "id": "reflection-beam-1-0-high",
    "displayName": "Beam 1.0 (High)",
    "modelReleaseDate": "2026-10-16T00:00:00.000Z",
    "providerId": "Reflection",
    "modelType": "CoT",
    "modelGroup": "reflection-beam-1-0",
    "paperUrl": "",
    "modelCardUrl": "",
    "dataUrl": "https://huggingface.co/reflection/Beam-1.0-501B-A23B",
    "featured": null,
    "notes": "Evaluation conducted with 'high' reasoning effort using the standard harness."
  },
  {
    "id": "reflection-beam-1-0-medium",
    "displayName": "Beam 1.0 (Medium)",
    "modelReleaseDate": "2026-10-16T00:00:00.000Z",
    "providerId": "Reflection",
    "modelType": "CoT",
    "modelGroup": "reflection-beam-1-0",
    "paperUrl": "",
    "modelCardUrl": "",
    "dataUrl": "https://huggingface.co/reflection/Beam-1.0-501B-A23B",
    "featured": null,
    "notes": "Evaluation conducted with 'medium' reasoning effort using the standard harness."
  },
  {
    "id": "reflection-beam-1-0-low",
    "displayName": "Beam 1.0 (Low)",
    "modelReleaseDate": "2026-10-16T00:00:00.000Z",
    "providerId": "Reflection",
    "modelType": "CoT",
    "modelGroup": "reflection-beam-1-0",
    "paperUrl": "",
    "modelCardUrl": "",
    "dataUrl": "https://huggingface.co/reflection/Beam-1.0-501B-A23B",
    "featured": null,
    "notes": "Evaluation conducted with 'low' reasoning effort using the standard harness."
  },
  {
    "id": "meta-muse-spark-open-xhigh",
    "displayName": "Muse Spark Open (XHigh)",
    "modelReleaseDate": "2026-10-15T00:00:00.000Z",
    "providerId": "Meta AI",
    "modelType": "CoT",
    "modelGroup": "meta-muse-spark-open",
    "paperUrl": "",
    "modelCardUrl": "",
    "dataUrl": "https://huggingface.co/meta-models/Muse-Spark-Open-816B-A42B",
    "featured": null,
    "notes": "Evaluation conducted with 'xhigh' reasoning effort using the standard harness."
  },
  {
    "id": "meta-muse-spark-open-high",
    "displayName": "Muse Spark Open (High)",
    "modelReleaseDate": "2026-10-15T00:00:00.000Z",
    "providerId": "Meta AI",
    "modelType": "CoT",
    "modelGroup": "meta-muse-spark-open",
    "paperUrl": "",
    "modelCardUrl": "",
    "dataUrl": "https://huggingface.co/meta-models/Muse-Spark-Open-816B-A42B",
    "featured": null,
    "notes": "Evaluation conducted with 'high' reasoning effort using the standard harness."
  },
  {
    "id": "meta-muse-spark-open-medium",
    "displayName": "Muse Spark Open (Medium)",
    "modelReleaseDate": "2026-10-15T00:00:00.000Z",
    "providerId": "Meta AI",
    "modelType": "CoT",
    "modelGroup": "meta-muse-spark-open",
    "paperUrl": "",
    "modelCardUrl": "",
    "dataUrl": "https://huggingface.co/meta-models/Muse-Spark-Open-816B-A42B",
    "featured": null,
    "notes": "Evaluation conducted with 'medium' reasoning effort using the standard harness."
  },
  {
    "id": "meta-muse-spark-open-low",
    "displayName": "Muse Spark Open (Low)",
    "modelReleaseDate": "2026-10-15T00:00:00.000Z",
    "providerId": "Meta AI",
    "modelType": "CoT",
    "modelGroup": "meta-muse-spark-open",
    "paperUrl": "",
    "modelCardUrl": "",
    "dataUrl": "https://huggingface.co/meta-models/Muse-Spark-Open-816B-A42B",
    "featured": null,
    "notes": "Evaluation conducted with 'low' reasoning effort using the standard harness."
  },
  {
    "id": "nvidia-nemotron-3-5-super-vl-120b-a12b-high",
    "displayName": "Nemotron 3.5 Super VL (High)",
    "modelReleaseDate": "2026-10-15T00:00:00.000Z",
    "providerId": "NVIDIA",
    "modelType": "CoT",
    "modelGroup": "nvidia-nemotron-3-5-super-vl-120b-a12b",
    "paperUrl": "",
    "modelCardUrl": "",
    "dataUrl": "https://huggingface.co/nvidia/NVIDIA-Nemotron-3.5-Super-VL-120B-A12B-BF16",
    "featured": null,
    "notes": "Evaluation conducted with 'high' reasoning effort using the standard harness."
  },
  {
    "id": "nvidia-nemotron-3-5-super-vl-120b-a12b-medium",
    "displayName": "Nemotron 3.5 Super VL (Medium)",
    "modelReleaseDate": "2026-10-15T00:00:00.000Z",
    "providerId": "NVIDIA",
    "modelType": "CoT",
    "modelGroup": "nvidia-nemotron-3-5-super-vl-120b-a12b",
    "paperUrl": "",
    "modelCardUrl": "",
    "dataUrl": "https://huggingface.co/nvidia/NVIDIA-Nemotron-3.5-Super-VL-120B-A12B-BF16",
    "featured": null,
    "notes": "Evaluation conducted with 'medium' reasoning effort using the standard harness. Partial dataset coverage: scores use full catalog denominators; costs include recorded usage only, excluding unrecorded attempts."
  },
  {
    "id": "nvidia-nemotron-3-5-super-vl-120b-a12b-none",
    "displayName": "Nemotron 3.5 Super VL (None)",
    "modelReleaseDate": "2026-10-15T00:00:00.000Z",
    "providerId": "NVIDIA",
    "modelType": "CoT",
    "modelGroup": "nvidia-nemotron-3-5-super-vl-120b-a12b",
    "paperUrl": "",
    "modelCardUrl": "",
    "dataUrl": "https://huggingface.co/nvidia/NVIDIA-Nemotron-3.5-Super-VL-120B-A12B-BF16",
    "featured": null,
    "notes": "Evaluation conducted with 'none' reasoning effort using the standard harness. Partial dataset coverage: scores use full catalog denominators; costs include recorded usage only, excluding unrecorded attempts."
  },
  {
    "id": "deepseek-v4-1-flash-1017-max",
    "displayName": "DeepSeek V4.1 Flash 1017 (Max)",
    "modelReleaseDate": "2026-10-17T00:00:00.000Z",
    "providerId": "DeepSeek",
    "modelType": "CoT",
    "modelGroup": "deepseek-v4-1-flash-1017",
    "paperUrl": "",
    "modelCardUrl": "",
    "dataUrl": "https://huggingface.co/deepseek-ai/DeepSeek-V4.1-Flash-1017",
    "featured": null,
    "notes": "Evaluation conducted with 'max' reasoning effort using the standard harness."
  },
  {
    "id": "deepseek-v4-1-flash-1017-high",
    "displayName": "DeepSeek V4.1 Flash 1017 (High)",
    "modelReleaseDate": "2026-10-17T00:00:00.000Z",
    "providerId": "DeepSeek",
    "modelType": "CoT",
    "modelGroup": "deepseek-v4-1-flash-1017",
    "paperUrl": "",
    "modelCardUrl": "",
    "dataUrl": "https://huggingface.co/deepseek-ai/DeepSeek-V4.1-Flash-1017",
    "featured": null,
    "notes": "Evaluation conducted with 'high' reasoning effort using the standard harness. Partial dataset coverage: scores use full catalog denominators; costs include recorded usage only, excluding unrecorded attempts."
  },
  {
    "id": "deepseek-v4-1-flash-1017-low",
    "displayName": "DeepSeek V4.1 Flash 1017 (Low)",
    "modelReleaseDate": "2026-10-17T00:00:00.000Z",
    "providerId": "DeepSeek",
    "modelType": "CoT",
    "modelGroup": "deepseek-v4-1-flash-1017",
    "paperUrl": "",
    "modelCardUrl": "",
    "dataUrl": "https://huggingface.co/deepseek-ai/DeepSeek-V4.1-Flash-1017",
    "featured": null,
    "notes": "Evaluation conducted with 'low' reasoning effort using the standard harness."
  },
  {
    "id": "deepseek-v4-1-flash-1017-none",
    "displayName": "DeepSeek V4.1 Flash 1017 (None)",
    "modelReleaseDate": "2026-10-17T00:00:00.000Z",
    "providerId": "DeepSeek",
    "modelType": "CoT",
    "modelGroup": "deepseek-v4-1-flash-1017",
    "paperUrl": "",
    "modelCardUrl": "",
    "dataUrl": "https://huggingface.co/deepseek-ai/DeepSeek-V4.1-Flash-1017",
    "featured": null,
    "notes": "Evaluation conducted with 'none' reasoning effort using the standard harness."
  },
  {
    "id": "anthropic-claude-sonnet-5-5-max",
    "displayName": "Claude Sonnet 5.5 (Max)",
    "modelReleaseDate": "2026-09-28T00:00:00.000Z",
    "providerId": "Anthropic",
    "modelType": "CoT",
    "modelGroup": "anthropic-claude-sonnet-5-5",
    "paperUrl": "",
    "modelCardUrl": "",
    "dataUrl": "",
    "featured": null,
    "notes": "Evaluation conducted with 'max' reasoning effort using the standard harness. Partial dataset coverage: scores use full catalog denominators; costs include recorded usage only, excluding unrecorded attempts."
  },
  {
    "id": "anthropic-claude-sonnet-5-5-xhigh",
    "displayName": "Claude Sonnet 5.5 (XHigh)",
    "modelReleaseDate": "2026-09-28T00:00:00.000Z",
    "providerId": "Anthropic",
    "modelType": "CoT",
    "modelGroup": "anthropic-claude-sonnet-5-5",
    "paperUrl": "",
    "modelCardUrl": "",
    "dataUrl": "",
    "featured": null,
    "notes": "Evaluation conducted with 'xhigh' reasoning effort using the standard harness."
  },
  {
    "id": "anthropic-claude-sonnet-5-5-high",
    "displayName": "Claude Sonnet 5.5 (High)",
    "modelReleaseDate": "2026-09-28T00:00:00.000Z",
    "providerId": "Anthropic",
    "modelType": "CoT",
    "modelGroup": "anthropic-claude-sonnet-5-5",
    "paperUrl": "",
    "modelCardUrl": "",
    "dataUrl": "",
    "featured": null,
    "notes": "Evaluation conducted with 'high' reasoning effort using the standard harness."
  },
  {
    "id": "anthropic-claude-sonnet-5-5-medium",
    "displayName": "Claude Sonnet 5.5 (Medium)",
    "modelReleaseDate": "2026-09-28T00:00:00.000Z",
    "providerId": "Anthropic",
    "modelType": "CoT",
    "modelGroup": "anthropic-claude-sonnet-5-5",
    "paperUrl": "",
    "modelCardUrl": "",
    "dataUrl": "",
    "featured": null,
    "notes": "Evaluation conducted with 'medium' reasoning effort using the standard harness."
  },
  {
    "id": "anthropic-claude-sonnet-5-5-low",
    "displayName": "Claude Sonnet 5.5 (Low)",
    "modelReleaseDate": "2026-09-28T00:00:00.000Z",
    "providerId": "Anthropic",
    "modelType": "CoT",
    "modelGroup": "anthropic-claude-sonnet-5-5",
    "paperUrl": "",
    "modelCardUrl": "",
    "dataUrl": "",
    "featured": null,
    "notes": "Evaluation conducted with 'low' reasoning effort using the standard harness. Partial dataset coverage: scores use full catalog denominators; costs include recorded usage only, excluding unrecorded attempts. Some model actions have unmeasured token usage; reported costs are recorded-usage lower bounds, not verified full-run costs."
  },
  {
    "id": "anthropic-claude-sonnet-5-5-xhigh-provider-adapter",
    "displayName": "Claude Sonnet 5.5 - Provider Adapter (XHigh)",
    "modelReleaseDate": "2026-09-28T00:00:00.000Z",
    "providerId": "Anthropic",
    "modelType": "CoT",
    "modelGroup": "anthropic-claude-sonnet-5-5-provider-adapter",
    "paperUrl": "",
    "modelCardUrl": "",
    "dataUrl": "",
    "featured": null,
    "notes": "Evaluation conducted with 'xhigh' reasoning effort using the provider adapter harness. Partial dataset coverage: scores use full catalog denominators; costs include recorded usage only, excluding unrecorded attempts."
  },
  {
    "id": "anthropic-claude-sonnet-5-5-high-provider-adapter",
    "displayName": "Claude Sonnet 5.5 - Provider Adapter (High)",
    "modelReleaseDate": "2026-09-28T00:00:00.000Z",
    "providerId": "Anthropic",
    "modelType": "CoT",
    "modelGroup": "anthropic-claude-sonnet-5-5-provider-adapter",
    "paperUrl": "",
    "modelCardUrl": "",
    "dataUrl": "",
    "featured": null,
    "notes": "Evaluation conducted with 'high' reasoning effort using the provider adapter harness. Partial dataset coverage: scores use full catalog denominators; costs include recorded usage only, excluding unrecorded attempts."
  },
  {
    "id": "anthropic-claude-sonnet-5-5-medium-provider-adapter",
    "displayName": "Claude Sonnet 5.5 - Provider Adapter (Medium)",
    "modelReleaseDate": "2026-09-28T00:00:00.000Z",
    "providerId": "Anthropic",
    "modelType": "CoT",
    "modelGroup": "anthropic-claude-sonnet-5-5-provider-adapter",
    "paperUrl": "",
    "modelCardUrl": "",
    "dataUrl": "",
    "featured": null,
    "notes": "Evaluation conducted with 'medium' reasoning effort using the provider adapter harness. Partial dataset coverage: scores use full catalog denominators; costs include recorded usage only, excluding unrecorded attempts."
  },
  {
    "id": "anthropic-claude-sonnet-5-5-low-provider-adapter",
    "displayName": "Claude Sonnet 5.5 - Provider Adapter (Low)",
    "modelReleaseDate": "2026-09-28T00:00:00.000Z",
    "providerId": "Anthropic",
    "modelType": "CoT",
    "modelGroup": "anthropic-claude-sonnet-5-5-provider-adapter",
    "paperUrl": "",
    "modelCardUrl": "",
    "dataUrl": "",
    "featured": null,
    "notes": "Evaluation conducted with 'low' reasoning effort using the provider adapter harness."
  }
]