{
  "api_version": "1",
  "updated": "2026-10-29",
  "latest_event": "2026-10-28",
  "rights": "© Frontier Log, all rights reserved.",
  "story": {
    "slug": "gemini-4-carbon-aa-scores",
    "title": "Gemini 4 Carbon's Table Moves Rivals' Scores",
    "description": "Google's Gemini 4 Carbon launch took its Terminal-Bench 4.0 scores from Artificial Analysis. Claude Opus 5.5 came out 6.8 points lower than a month before.",
    "url": "https://frontierlog.ai/stories/gemini-4-carbon-aa-scores",
    "date": "2026-10-29",
    "image": "https://frontierlog.ai/og/story-gemini-4-carbon-aa-scores.png",
    "event": "2026-10-28-gemini-4-carbon",
    "tags": [
      "coding",
      "agents"
    ],
    "steps": [],
    "tables": [
      {
        "title": "Terminal-Bench 4.0, each lab's own report",
        "unit": "%",
        "rows": [
          {
            "event": "2026-09-30-gemini-4-argon",
            "date": "2026-09-30",
            "provider": "google",
            "value": 57.4,
            "printed": "57.4%",
            "source": "https://blog.google/innovation-and-ai/models-and-research/gemini-models/gemini-4-argon/",
            "where": "benchmark table (image), row 'Terminal-bench 4.0', column 'Gemini 4 Argon'"
          },
          {
            "event": "2026-09-03-gpt-6-astra",
            "date": "2026-09-03",
            "provider": "openai",
            "value": 57.9,
            "printed": "57.9%",
            "source": "https://openai.com/index/gpt-6-astra",
            "where": "Coding table, row 'Terminal-Bench 4.0', column 'GPT‑6 Astra'"
          },
          {
            "event": "2026-09-22-claude-opus-5-5",
            "date": "2026-09-22",
            "provider": "anthropic",
            "value": 66.4,
            "printed": "66.4%",
            "source": "https://www.anthropic.com/claude-opus-5-5",
            "where": "benchmark table, row 'Terminal-Bench 4.0', column 'Opus 5.5'"
          },
          {
            "event": "2026-09-01-claude-fable-5-1-mythos-5-1",
            "date": "2026-09-01",
            "provider": "anthropic",
            "value": 55.8,
            "printed": "55.8%",
            "source": "https://www.anthropic.com/claude-fable-and-mythos-5-1",
            "where": "benchmark table, row 'Terminal-Bench 4.0', column 'Fable 5.1' (the cell also prints a Mythos 5.1 figure)"
          }
        ]
      },
      {
        "title": "Terminal-Bench 4.0, in Google's Gemini 4 Argon table",
        "unit": "%",
        "rows": [
          {
            "event": "2026-09-30-gemini-4-argon",
            "date": "2026-09-30",
            "provider": "google",
            "value": 57.4,
            "printed": null,
            "source": "https://blog.google/innovation-and-ai/models-and-research/gemini-models/gemini-4-argon/",
            "where": "benchmark table (image), row 'Terminal-bench 4.0', column 'Gemini 4 Argon'"
          },
          {
            "event": "2026-09-30-gemini-4-argon",
            "date": "2026-09-30",
            "provider": "google",
            "value": 58.2,
            "printed": null,
            "source": "https://blog.google/innovation-and-ai/models-and-research/gemini-models/gemini-4-argon/",
            "where": "benchmark table (image), row 'Terminal-bench 4.0', column 'GPT-6 Astra'"
          },
          {
            "event": "2026-09-30-gemini-4-argon",
            "date": "2026-09-30",
            "provider": "google",
            "value": 66.4,
            "printed": null,
            "source": "https://blog.google/innovation-and-ai/models-and-research/gemini-models/gemini-4-argon/",
            "where": "benchmark table (image), row 'Terminal-bench 4.0', column 'Claude Opus 5.5'"
          },
          {
            "event": "2026-09-30-gemini-4-argon",
            "date": "2026-09-30",
            "provider": "google",
            "value": 57.9,
            "printed": null,
            "source": "https://blog.google/innovation-and-ai/models-and-research/gemini-models/gemini-4-argon/",
            "where": "benchmark table (image), row 'Terminal-bench 4.0', column 'Claude Fable 5.1'"
          }
        ]
      },
      {
        "title": "Terminal-Bench 4.0, in Google's Gemini 4 Carbon table, from Artificial Analysis",
        "unit": "%",
        "rows": [
          {
            "event": "2026-10-28-gemini-4-carbon",
            "date": "2026-10-28",
            "provider": "google",
            "value": 65.2,
            "printed": null,
            "source": "https://blog.google/innovation-and-ai/models-and-research/gemini-models/introducing-gemini-4-carbon/",
            "where": "benchmark table (image), row 'Terminal-Bench 4.0 (AA)', column 'Gemini 4 Carbon'"
          },
          {
            "event": "2026-10-28-gemini-4-carbon",
            "date": "2026-10-28",
            "provider": "google",
            "value": 57.1,
            "printed": null,
            "source": "https://blog.google/innovation-and-ai/models-and-research/gemini-models/introducing-gemini-4-carbon/",
            "where": "benchmark table (image), row 'Terminal-Bench 4.0 (AA)', column 'Gemini 4 Argon'"
          },
          {
            "event": "2026-10-28-gemini-4-carbon",
            "date": "2026-10-28",
            "provider": "google",
            "value": 59.1,
            "printed": null,
            "source": "https://blog.google/innovation-and-ai/models-and-research/gemini-models/introducing-gemini-4-carbon/",
            "where": "benchmark table (image), row 'Terminal-Bench 4.0 (AA)', column 'GPT-6 Astra'"
          },
          {
            "event": "2026-10-28-gemini-4-carbon",
            "date": "2026-10-28",
            "provider": "google",
            "value": 59.6,
            "printed": null,
            "source": "https://blog.google/innovation-and-ai/models-and-research/gemini-models/introducing-gemini-4-carbon/",
            "where": "benchmark table (image), row 'Terminal-Bench 4.0 (AA)', column 'Claude Opus 5.5'"
          },
          {
            "event": "2026-10-28-gemini-4-carbon",
            "date": "2026-10-28",
            "provider": "google",
            "value": 52.0,
            "printed": null,
            "source": "https://blog.google/innovation-and-ai/models-and-research/gemini-models/introducing-gemini-4-carbon/",
            "where": "benchmark table (image), row 'Terminal-Bench 4.0 (AA)', column 'Claude Fable 5.1'"
          }
        ]
      }
    ],
    "records": [
      {
        "date": "2025-10-27",
        "party": "MiniMax",
        "what": "MiniMax-M2's model card prints every model's scores from Artificial Analysis, under the heading 'Metric (AA)'",
        "by": "MiniMax",
        "source": "https://huggingface.co/MiniMaxAI/MiniMax-M2/blob/44cefa66f81ec7ffebcb580c02985726ef4d829e/README.md",
        "quote": "All scores of other models reported from https://artificialanalysis.ai/."
      },
      {
        "date": "2026-09-02",
        "party": "Google",
        "what": "Gemini 3.8 Flash's table takes other models' Terminal-Bench 2.1 scores from the leaderboard and Artificial Analysis",
        "by": "Google",
        "source": "https://storage.googleapis.com/deepmind-media/gemini/gemini_3-8_flash_model_evaluation.pdf",
        "quote": "Terminal-Bench 2.1 results for Gemini models are self computed and for other models are reported from the public leaderboard and Artificial Analysis."
      },
      {
        "date": "2026-09-30",
        "party": "Google",
        "what": "Gemini 4 Argon's table computes Argon's Terminal-Bench scores itself and takes other models' from the leaderboard",
        "by": "Google",
        "source": "https://storage.googleapis.com/deepmind-media/gemini/gemini_4_argon_model_evaluation.pdf",
        "quote": "Terminal-Bench 4.0 results for Gemini 4 Argon are self computed, all other models are reported from the official public leaderboard."
      },
      {
        "date": "2026-10-06",
        "party": "Mistral AI",
        "what": "Mistral Large 4's Terminal-Bench 4.0 chart is labelled 'AA - Terminal-Bench 4'",
        "by": "Mistral AI",
        "source": "https://mistral.ai/news/mistral-large-4/",
        "quote": "Scores use the numbers evaluated privately by Artificial Analysis ahead of the harness' public launch"
      },
      {
        "date": "2026-10-28",
        "party": "Google",
        "what": "Gemini 4 Carbon's table takes both Terminal-Bench rows from Artificial Analysis for every model",
        "by": "Google",
        "source": "https://storage.googleapis.com/deepmind-media/gemini/gemini_4_carbon_model_evaluation.pdf",
        "quote": "Terminal-Bench 4.0 (AA) results are sourced from Artificial Analysis.",
        "key": true
      }
    ],
    "cites": [
      {
        "url": "https://huggingface.co/MiniMaxAI/MiniMax-M2/blob/44cefa66f81ec7ffebcb580c02985726ef4d829e/README.md",
        "by": "MiniMax",
        "medium": "model card, launch-day revision",
        "date": "2025-10-27"
      },
      {
        "url": "https://storage.googleapis.com/deepmind-media/gemini/gemini_3-8_flash_model_evaluation.pdf",
        "by": "Google",
        "medium": "model evaluation document",
        "date": "2026-09-02"
      },
      {
        "url": "https://storage.googleapis.com/deepmind-media/gemini/gemini_4_argon_model_evaluation.pdf",
        "by": "Google",
        "medium": "model evaluation document",
        "date": "2026-09-30"
      },
      {
        "url": "https://mistral.ai/news/mistral-large-4/",
        "by": "Mistral AI",
        "medium": "blog post",
        "date": "2026-10-06"
      },
      {
        "url": "https://storage.googleapis.com/deepmind-media/gemini/gemini_4_carbon_model_evaluation.pdf",
        "by": "Google",
        "medium": "model evaluation document",
        "date": "2026-10-28"
      }
    ],
    "sources": [
      "https://blog.google/innovation-and-ai/models-and-research/gemini-models/introducing-gemini-4-carbon/",
      "https://storage.googleapis.com/deepmind-media/gemini/gemini_4_carbon_model_evaluation.pdf",
      "https://blog.google/innovation-and-ai/models-and-research/gemini-models/gemini-4-argon/",
      "https://storage.googleapis.com/deepmind-media/gemini/gemini_4_argon_model_evaluation.pdf",
      "https://blog.google/innovation-and-ai/models-and-research/gemini-models/3-8-flash-and-3-8-flash-cyber/",
      "https://storage.googleapis.com/deepmind-media/gemini/gemini_3-8_flash_model_evaluation.pdf",
      "https://www.anthropic.com/claude-opus-5-5",
      "https://www.anthropic.com/claude-fable-and-mythos-5-1",
      "https://openai.com/index/gpt-6-astra",
      "https://mistral.ai/news/mistral-large-4/",
      "https://huggingface.co/MiniMaxAI/MiniMax-M2/blob/44cefa66f81ec7ffebcb580c02985726ef4d829e/README.md"
    ],
    "body": [
      {
        "figure": "main"
      },
      {
        "p": "Google launched Gemini 4 Carbon [on Oct 28](https://blog.google/innovation-and-ai/models-and-research/gemini-models/introducing-gemini-4-carbon/) with a table comparing it with Gemini 4 Argon, GPT-6 Astra, Claude Opus 5.5 and Claude Fable 5.1. Its Terminal-Bench 4.0 row is marked \"(AA)\": Google's [evaluation document](https://storage.googleapis.com/deepmind-media/gemini/gemini_4_carbon_model_evaluation.pdf) says the results \"are sourced from Artificial Analysis\" for every model, Carbon included. [Argon’s launch table](https://blog.google/innovation-and-ai/models-and-research/gemini-models/gemini-4-argon/) printed the same row for the same four models on Sep 30, from other sources."
      },
      {
        "h": "Three figures for most models"
      },
      {
        "p": "For Claude Opus 5.5, [Anthropic reported](https://www.anthropic.com/claude-opus-5-5) 66.4% at its \"xhigh\" effort setting, and Argon's table printed the same 66.4%. Carbon's table prints 59.6%, 6.8 points lower. For Claude Fable 5.1, [Anthropic reported](https://www.anthropic.com/claude-fable-and-mythos-5-1) 55.8%, Argon's table printed 57.9% and Carbon's prints 52.0%."
      },
      {
        "p": "GPT-6 Astra moves the other way: [OpenAI reported](https://openai.com/index/gpt-6-astra) 57.9%, Argon's table printed 58.2% and Carbon's prints 59.1%. Google's own Argon scored 57.4% in Google's run at its launch and 57.1% in Artificial Analysis's figure in Carbon's table."
      },
      {
        "h": "Where each table's figures came from"
      },
      {
        "p": "Argon's [evaluation document](https://storage.googleapis.com/deepmind-media/gemini/gemini_4_argon_model_evaluation.pdf) says Google computed Argon's score itself and took the others \"from the official public leaderboard\", using \"the highest scoring thinking level for each model as reported by the Terminal Bench authors\". Carbon's document gives no harness or effort setting for the Artificial Analysis figures."
      },
      {
        "p": "Carbon's own 65.2% is also Artificial Analysis's figure, and it is the top score in Carbon's row. In Argon's table, the top score was Opus 5.5's 66.4%."
      },
      {
        "p": "For Gemini 3.8 Flash [on Sep 2](https://blog.google/innovation-and-ai/models-and-research/gemini-models/3-8-flash-and-3-8-flash-cyber/), Google took other models' Terminal-Bench 2.1 scores \"from the public leaderboard and Artificial Analysis\", its [evaluation document](https://storage.googleapis.com/deepmind-media/gemini/gemini_3-8_flash_model_evaluation.pdf) says, without marking which came from where."
      },
      {
        "h": "Two other launches printed AA's runs"
      },
      {
        "p": "Of the 70 flagship launch posts in Frontier Log's reporting inventory, from January 2025 to mid-October 2026, two printed Artificial Analysis's runs of benchmarks made by others. [MiniMax-M2’s model card](https://huggingface.co/MiniMaxAI/MiniMax-M2/blob/44cefa66f81ec7ffebcb580c02985726ef4d829e/README.md) in October 2025 headed its table \"Metric (AA)\" and said all other models' scores came from Artificial Analysis. [Mistral Large 4](https://mistral.ai/news/mistral-large-4/), on Oct 6, labelled its Terminal-Bench 4.0 chart \"AA - Terminal-Bench 4\". Benchmarks that Artificial Analysis builds itself, such as GDPval-AA, appear far more often."
      },
      {
        "p": "For how much the agent harness alone can move one model's score, see [Does the Harness Version Change a Model's Score?](https://frontierlog.ai/stories/does-harness-version-change-score)."
      }
    ],
    "models": [
      "gemini-4-carbon",
      "gemini-4-argon",
      "gpt-6-astra",
      "claude-opus-5-5",
      "claude-fable-5-1"
    ],
    "stories": [
      "does-harness-version-change-score"
    ]
  }
}