{
  "schemaVersion": 1,
  "benchmark": {
    "id": "terminal-bench-science",
    "name": "Terminal-Bench-Science",
    "version": "0.1.0",
    "category": "science",
    "question": "Can an agent carry out scientific work?",
    "summary": "Research tasks that require scientific software, computation, and evidence across five domains.",
    "source": {
      "name": "Terminal-Bench-Science",
      "url": "https://hub.harborframework.com/datasets/terminal-bench-science/terminal-bench-science/0.1.0?leaderboard=v0-1-eval&tab=leaderboard",
      "methodologyUrl": "https://www.terminal-bench-science.ai/announcement"
    },
    "measure": "Resolution rate across 70 tasks and three trials, with separate domain results.",
    "comparisonRule": "Compare the exact 0.1.0 cohort. Error bars show one standard error, which is narrower than a 95% confidence interval.",
    "limitations": [
      "Model and agent versions and several run limits are not specified by the source.",
      "Cost covers the full 210-trial evaluation; source aggregate and domain costs can differ.",
      "A benchmark success does not establish the scientific validity of arbitrary new research."
    ],
    "coverage": "charted",
    "tags": [
      "scientific computing",
      "research",
      "engineering",
      "life science",
      "physics",
      "mathematics"
    ]
  },
  "dataset": {
    "benchmarkId": "terminal-bench-science",
    "version": "0.1.0",
    "score": {
      "label": "Resolution rate",
      "unit": "%",
      "direction": "higher",
      "minimum": 0,
      "maximum": 100
    },
    "source": {
      "name": "Terminal-Bench-Science and Harbor Framework",
      "url": "https://hub.harborframework.com/datasets/terminal-bench-science/terminal-bench-science/0.1.0?leaderboard=v0-1-eval&tab=leaderboard",
      "retrievedAt": "2026-09-25T15:34:56.683Z",
      "revision": "f81afac4f11048e77a15dfc8fb1dbfb897fea0ce"
    },
    "observedAt": "2026-09-23T16:29:34.906395+00:00",
    "evidenceLabel": "Owner leaderboard",
    "configurationLabel": "Model, coding agent, and effort",
    "comparabilityNote": "70 science tasks × 3 trials. Error bars show one binomial standard error, not a 95% confidence interval. Model and agent versions, price basis, and several run limits are unspecified.",
    "costLabel": "USD for the full 210-trial evaluation",
    "points": [
      {
        "id": "76867ebc-8f35-43c5-a8b0-0cee9db46c96",
        "label": "GPT-6 Astra · Codex · max",
        "model": "GPT-6 Astra",
        "provider": "OpenAI",
        "harness": "Codex",
        "effort": "max",
        "score": 68.0952380952381,
        "costUsd": 4997.461181,
        "uncertainty": {
          "lower": 64.87879051453473,
          "upper": 71.31168567594148,
          "label": "Source-reported ±1 binomial standard error"
        },
        "sourceUrl": "https://hub.harborframework.com/datasets/terminal-bench-science/terminal-bench-science/0.1.0/leaderboards/v0-1-eval/rows/76867ebc-8f35-43c5-a8b0-0cee9db46c96",
        "details": [
          {
            "label": "Earth science",
            "value": "79.2% · 24 trials"
          },
          {
            "label": "Engineering science",
            "value": "70.4% · 27 trials"
          },
          {
            "label": "Life science",
            "value": "59.6% · 57 trials"
          },
          {
            "label": "Mathematical science",
            "value": "80.4% · 51 trials"
          },
          {
            "label": "Physical science",
            "value": "58.8% · 51 trials"
          }
        ]
      },
      {
        "id": "cb24f3a2-82ab-4f97-be88-92a28118bf1c",
        "label": "Opus 5.5 · Claude Code · max",
        "model": "Opus 5.5",
        "provider": "Anthropic",
        "harness": "Claude Code",
        "effort": "max",
        "score": 63.33333333333333,
        "costUsd": 4874.2576698,
        "uncertainty": {
          "lower": 60.007945978723185,
          "upper": 66.65872068794347,
          "label": "Source-reported ±1 binomial standard error"
        },
        "sourceUrl": "https://hub.harborframework.com/datasets/terminal-bench-science/terminal-bench-science/0.1.0/leaderboards/v0-1-eval/rows/cb24f3a2-82ab-4f97-be88-92a28118bf1c",
        "details": [
          {
            "label": "Earth science",
            "value": "62.5% · 24 trials"
          },
          {
            "label": "Engineering science",
            "value": "63.0% · 27 trials"
          },
          {
            "label": "Life science",
            "value": "54.4% · 57 trials"
          },
          {
            "label": "Mathematical science",
            "value": "76.5% · 51 trials"
          },
          {
            "label": "Physical science",
            "value": "60.8% · 51 trials"
          }
        ]
      },
      {
        "id": "1264a994-632f-4742-a143-a848a6ad4114",
        "label": "Fable 5.1 · Claude Code · max",
        "model": "Fable 5.1",
        "provider": "Anthropic",
        "harness": "Claude Code",
        "effort": "max",
        "score": 40,
        "costUsd": 6247.1398385,
        "uncertainty": {
          "lower": 36.619382981085934,
          "upper": 43.380617018914066,
          "label": "Source-reported ±1 binomial standard error"
        },
        "sourceUrl": "https://hub.harborframework.com/datasets/terminal-bench-science/terminal-bench-science/0.1.0/leaderboards/v0-1-eval/rows/1264a994-632f-4742-a143-a848a6ad4114",
        "details": [
          {
            "label": "Earth science",
            "value": "37.5% · 24 trials"
          },
          {
            "label": "Engineering science",
            "value": "48.1% · 27 trials"
          },
          {
            "label": "Life science",
            "value": "38.6% · 57 trials"
          },
          {
            "label": "Mathematical science",
            "value": "41.2% · 51 trials"
          },
          {
            "label": "Physical science",
            "value": "37.3% · 51 trials"
          }
        ]
      },
      {
        "id": "bcf50bd8-8555-48a6-8cae-9f697af54051",
        "label": "Opus 5 · Claude Code · max",
        "model": "Opus 5",
        "provider": "Anthropic",
        "harness": "Claude Code",
        "effort": "max",
        "score": 30,
        "costUsd": 6992.67732375,
        "uncertainty": {
          "lower": 26.837722339831622,
          "upper": 33.16227766016838,
          "label": "Source-reported ±1 binomial standard error"
        },
        "sourceUrl": "https://hub.harborframework.com/datasets/terminal-bench-science/terminal-bench-science/0.1.0/leaderboards/v0-1-eval/rows/bcf50bd8-8555-48a6-8cae-9f697af54051",
        "details": [
          {
            "label": "Earth science",
            "value": "45.8% · 24 trials"
          },
          {
            "label": "Engineering science",
            "value": "29.6% · 27 trials"
          },
          {
            "label": "Life science",
            "value": "29.8% · 57 trials"
          },
          {
            "label": "Mathematical science",
            "value": "25.5% · 51 trials"
          },
          {
            "label": "Physical science",
            "value": "27.5% · 51 trials"
          }
        ]
      },
      {
        "id": "378f8b49-dbd8-4acf-a065-8c07e1712de2",
        "label": "GPT-5.6 Sol · Codex · max",
        "model": "GPT-5.6 Sol",
        "provider": "OpenAI",
        "harness": "Codex",
        "effort": "max",
        "score": 22.380952380952383,
        "costUsd": 4221.8683028,
        "uncertainty": {
          "lower": 19.50478743385059,
          "upper": 25.257117328054175,
          "label": "Source-reported ±1 binomial standard error"
        },
        "sourceUrl": "https://hub.harborframework.com/datasets/terminal-bench-science/terminal-bench-science/0.1.0/leaderboards/v0-1-eval/rows/378f8b49-dbd8-4acf-a065-8c07e1712de2",
        "details": [
          {
            "label": "Earth science",
            "value": "20.8% · 24 trials"
          },
          {
            "label": "Engineering science",
            "value": "14.8% · 27 trials"
          },
          {
            "label": "Life science",
            "value": "17.5% · 57 trials"
          },
          {
            "label": "Mathematical science",
            "value": "31.4% · 51 trials"
          },
          {
            "label": "Physical science",
            "value": "23.5% · 51 trials"
          }
        ]
      },
      {
        "id": "e3f2861a-97db-4730-a0ff-99f08d8c7299",
        "label": "Fable 5 · Claude Code · max",
        "model": "Fable 5",
        "provider": "Anthropic",
        "harness": "Claude Code",
        "effort": "max",
        "score": 21.428571428571427,
        "costUsd": 14175.948185250001,
        "uncertainty": {
          "lower": 18.5970536886711,
          "upper": 24.260089168471755,
          "label": "Source-reported ±1 binomial standard error"
        },
        "sourceUrl": "https://hub.harborframework.com/datasets/terminal-bench-science/terminal-bench-science/0.1.0/leaderboards/v0-1-eval/rows/e3f2861a-97db-4730-a0ff-99f08d8c7299",
        "details": [
          {
            "label": "Earth science",
            "value": "25.0% · 24 trials"
          },
          {
            "label": "Engineering science",
            "value": "11.1% · 27 trials"
          },
          {
            "label": "Life science",
            "value": "17.5% · 57 trials"
          },
          {
            "label": "Mathematical science",
            "value": "33.3% · 51 trials"
          },
          {
            "label": "Physical science",
            "value": "17.6% · 51 trials"
          }
        ]
      },
      {
        "id": "58d4c48d-791b-4873-ba3e-af1cf997a79c",
        "label": "DeepSeek V4.1 Flash · Codex · max",
        "model": "DeepSeek V4.1 Flash",
        "provider": "DeepSeek",
        "harness": "Codex",
        "effort": "max",
        "score": 15.714285714285714,
        "costUsd": 385.933474254,
        "uncertainty": {
          "lower": 13.202892820635272,
          "upper": 18.225678607936157,
          "label": "Source-reported ±1 binomial standard error"
        },
        "sourceUrl": "https://hub.harborframework.com/datasets/terminal-bench-science/terminal-bench-science/0.1.0/leaderboards/v0-1-eval/rows/58d4c48d-791b-4873-ba3e-af1cf997a79c",
        "details": [
          {
            "label": "Earth science",
            "value": "8.3% · 24 trials"
          },
          {
            "label": "Engineering science",
            "value": "18.5% · 27 trials"
          },
          {
            "label": "Life science",
            "value": "10.5% · 57 trials"
          },
          {
            "label": "Mathematical science",
            "value": "11.8% · 51 trials"
          },
          {
            "label": "Physical science",
            "value": "27.5% · 51 trials"
          }
        ]
      },
      {
        "id": "360317df-baea-4ead-8a11-09db70a27d15",
        "label": "Grok 4.7 · Grok Build · xhigh",
        "model": "Grok 4.7",
        "provider": "xAI",
        "harness": "Grok Build",
        "effort": "xhigh",
        "score": 14.285714285714286,
        "costUsd": 3070.816824,
        "uncertainty": {
          "lower": 11.870987843632811,
          "upper": 16.700440727795762,
          "label": "Source-reported ±1 binomial standard error"
        },
        "sourceUrl": "https://hub.harborframework.com/datasets/terminal-bench-science/terminal-bench-science/0.1.0/leaderboards/v0-1-eval/rows/360317df-baea-4ead-8a11-09db70a27d15",
        "details": [
          {
            "label": "Earth science",
            "value": "25.0% · 24 trials"
          },
          {
            "label": "Engineering science",
            "value": "25.9% · 27 trials"
          },
          {
            "label": "Life science",
            "value": "12.3% · 57 trials"
          },
          {
            "label": "Mathematical science",
            "value": "3.9% · 51 trials"
          },
          {
            "label": "Physical science",
            "value": "15.7% · 51 trials"
          }
        ]
      },
      {
        "id": "409f68a6-d6a4-43c9-adb8-63679dc17a45",
        "label": "Gemini 3.8 Flash · mini-SWE-agent · high",
        "model": "Gemini 3.8 Flash",
        "provider": "Google DeepMind",
        "harness": "mini-SWE-agent",
        "effort": "high",
        "score": 12.380952380952381,
        "costUsd": 1116.97645035,
        "uncertainty": {
          "lower": 10.108124002209669,
          "upper": 14.653780759695094,
          "label": "Source-reported ±1 binomial standard error"
        },
        "sourceUrl": "https://hub.harborframework.com/datasets/terminal-bench-science/terminal-bench-science/0.1.0/leaderboards/v0-1-eval/rows/409f68a6-d6a4-43c9-adb8-63679dc17a45",
        "details": [
          {
            "label": "Earth science",
            "value": "12.5% · 24 trials"
          },
          {
            "label": "Engineering science",
            "value": "11.1% · 27 trials"
          },
          {
            "label": "Life science",
            "value": "14.0% · 57 trials"
          },
          {
            "label": "Mathematical science",
            "value": "0.0% · 51 trials"
          },
          {
            "label": "Physical science",
            "value": "23.5% · 51 trials"
          }
        ]
      },
      {
        "id": "e8bf7739-3c94-46a3-95a5-fa4d4444034e",
        "label": "Opus 4.8 · Claude Code · max",
        "model": "Opus 4.8",
        "provider": "Anthropic",
        "harness": "Claude Code",
        "effort": "max",
        "score": 10.476190476190476,
        "costUsd": 5837.716905,
        "uncertainty": {
          "lower": 8.362889649424979,
          "upper": 12.589491302955974,
          "label": "Source-reported ±1 binomial standard error"
        },
        "sourceUrl": "https://hub.harborframework.com/datasets/terminal-bench-science/terminal-bench-science/0.1.0/leaderboards/v0-1-eval/rows/e8bf7739-3c94-46a3-95a5-fa4d4444034e",
        "details": [
          {
            "label": "Earth science",
            "value": "4.2% · 24 trials"
          },
          {
            "label": "Engineering science",
            "value": "7.4% · 27 trials"
          },
          {
            "label": "Life science",
            "value": "7.0% · 57 trials"
          },
          {
            "label": "Mathematical science",
            "value": "7.8% · 51 trials"
          },
          {
            "label": "Physical science",
            "value": "21.6% · 51 trials"
          }
        ]
      },
      {
        "id": "c79d5ef6-b3c6-40da-84b5-946c11099426",
        "label": "GPT-5.6 Terra · Codex · max",
        "model": "GPT-5.6 Terra",
        "provider": "OpenAI",
        "harness": "Codex",
        "effort": "max",
        "score": 8.571428571428571,
        "costUsd": 1983.9189368999998,
        "uncertainty": {
          "lower": 6.63964741776339,
          "upper": 10.503209725093752,
          "label": "Source-reported ±1 binomial standard error"
        },
        "sourceUrl": "https://hub.harborframework.com/datasets/terminal-bench-science/terminal-bench-science/0.1.0/leaderboards/v0-1-eval/rows/c79d5ef6-b3c6-40da-84b5-946c11099426",
        "details": [
          {
            "label": "Earth science",
            "value": "8.3% · 24 trials"
          },
          {
            "label": "Engineering science",
            "value": "7.4% · 27 trials"
          },
          {
            "label": "Life science",
            "value": "7.0% · 57 trials"
          },
          {
            "label": "Mathematical science",
            "value": "9.8% · 51 trials"
          },
          {
            "label": "Physical science",
            "value": "9.8% · 51 trials"
          }
        ]
      },
      {
        "id": "78f18248-3ee9-4eb4-b7af-19019f1fc64f",
        "label": "GLM 5.3 · Claude Code · max",
        "model": "GLM 5.3",
        "provider": "Z.AI",
        "harness": "Claude Code",
        "effort": "max",
        "score": 8.095238095238095,
        "costUsd": 2733.16185912,
        "uncertainty": {
          "lower": 6.213001672527808,
          "upper": 9.977474517948382,
          "label": "Source-reported ±1 binomial standard error"
        },
        "sourceUrl": "https://hub.harborframework.com/datasets/terminal-bench-science/terminal-bench-science/0.1.0/leaderboards/v0-1-eval/rows/78f18248-3ee9-4eb4-b7af-19019f1fc64f",
        "details": [
          {
            "label": "Earth science",
            "value": "4.2% · 24 trials"
          },
          {
            "label": "Engineering science",
            "value": "7.4% · 27 trials"
          },
          {
            "label": "Life science",
            "value": "10.5% · 57 trials"
          },
          {
            "label": "Mathematical science",
            "value": "5.9% · 51 trials"
          },
          {
            "label": "Physical science",
            "value": "9.8% · 51 trials"
          }
        ]
      },
      {
        "id": "66e64f64-abbd-40a6-a470-2151df25472e",
        "label": "Kimi K3 · Claude Code · max",
        "model": "Kimi K3",
        "provider": "Moonshot AI",
        "harness": "Claude Code",
        "effort": "max",
        "score": 7.142857142857142,
        "costUsd": 1523.7995976,
        "uncertainty": {
          "lower": 5.365666601685287,
          "upper": 8.920047684028997,
          "label": "Source-reported ±1 binomial standard error"
        },
        "sourceUrl": "https://hub.harborframework.com/datasets/terminal-bench-science/terminal-bench-science/0.1.0/leaderboards/v0-1-eval/rows/66e64f64-abbd-40a6-a470-2151df25472e",
        "details": [
          {
            "label": "Earth science",
            "value": "12.5% · 24 trials"
          },
          {
            "label": "Engineering science",
            "value": "3.7% · 27 trials"
          },
          {
            "label": "Life science",
            "value": "7.0% · 57 trials"
          },
          {
            "label": "Mathematical science",
            "value": "2.0% · 51 trials"
          },
          {
            "label": "Physical science",
            "value": "11.8% · 51 trials"
          }
        ]
      },
      {
        "id": "231bc6c3-8f2b-455e-8c87-5a08ec200b05",
        "label": "Grok 4.6 · Grok Build · xhigh",
        "model": "Grok 4.6",
        "provider": "xAI",
        "harness": "Grok Build",
        "effort": "xhigh",
        "score": 7.142857142857142,
        "costUsd": 3344.187332,
        "uncertainty": {
          "lower": 5.365666601685287,
          "upper": 8.920047684028997,
          "label": "Source-reported ±1 binomial standard error"
        },
        "sourceUrl": "https://hub.harborframework.com/datasets/terminal-bench-science/terminal-bench-science/0.1.0/leaderboards/v0-1-eval/rows/231bc6c3-8f2b-455e-8c87-5a08ec200b05",
        "details": [
          {
            "label": "Earth science",
            "value": "4.2% · 24 trials"
          },
          {
            "label": "Engineering science",
            "value": "14.8% · 27 trials"
          },
          {
            "label": "Life science",
            "value": "7.0% · 57 trials"
          },
          {
            "label": "Mathematical science",
            "value": "5.9% · 51 trials"
          },
          {
            "label": "Physical science",
            "value": "5.9% · 51 trials"
          }
        ]
      },
      {
        "id": "5a71d088-1bfd-4bbe-a65a-973a4fffd50f",
        "label": "Gemini 3.7 Flash · mini-SWE-agent · high",
        "model": "Gemini 3.7 Flash",
        "provider": "Google DeepMind",
        "harness": "mini-SWE-agent",
        "effort": "high",
        "score": 5.714285714285714,
        "costUsd": 1296.5871279,
        "uncertainty": {
          "lower": 4.112537398338891,
          "upper": 7.3160340302325375,
          "label": "Source-reported ±1 binomial standard error"
        },
        "sourceUrl": "https://hub.harborframework.com/datasets/terminal-bench-science/terminal-bench-science/0.1.0/leaderboards/v0-1-eval/rows/5a71d088-1bfd-4bbe-a65a-973a4fffd50f",
        "details": [
          {
            "label": "Earth science",
            "value": "0.0% · 24 trials"
          },
          {
            "label": "Engineering science",
            "value": "7.4% · 27 trials"
          },
          {
            "label": "Life science",
            "value": "8.8% · 57 trials"
          },
          {
            "label": "Mathematical science",
            "value": "2.0% · 51 trials"
          },
          {
            "label": "Physical science",
            "value": "7.8% · 51 trials"
          }
        ]
      },
      {
        "id": "00b5eabf-caca-435c-b8a8-458633192aa8",
        "label": "DeepSeek V4 Pro · Codex · max",
        "model": "DeepSeek V4 Pro",
        "provider": "DeepSeek",
        "harness": "Codex",
        "effort": "max",
        "score": 3.8095238095238093,
        "costUsd": 1120.15076932,
        "uncertainty": {
          "lower": 2.488557515845738,
          "upper": 5.130490103201881,
          "label": "Source-reported ±1 binomial standard error"
        },
        "sourceUrl": "https://hub.harborframework.com/datasets/terminal-bench-science/terminal-bench-science/0.1.0/leaderboards/v0-1-eval/rows/00b5eabf-caca-435c-b8a8-458633192aa8",
        "details": [
          {
            "label": "Earth science",
            "value": "0.0% · 24 trials"
          },
          {
            "label": "Engineering science",
            "value": "11.1% · 27 trials"
          },
          {
            "label": "Life science",
            "value": "0.0% · 57 trials"
          },
          {
            "label": "Mathematical science",
            "value": "5.9% · 51 trials"
          },
          {
            "label": "Physical science",
            "value": "3.9% · 51 trials"
          }
        ]
      },
      {
        "id": "7cf10b06-3e5b-4723-b98f-c679e773959b",
        "label": "GPT-5.6 Luna · Codex · max",
        "model": "GPT-5.6 Luna",
        "provider": "OpenAI",
        "harness": "Codex",
        "effort": "max",
        "score": 3.3333333333333335,
        "costUsd": 383.19963959999995,
        "uncertainty": {
          "lower": 2.094627745071323,
          "upper": 4.572038921595344,
          "label": "Source-reported ±1 binomial standard error"
        },
        "sourceUrl": "https://hub.harborframework.com/datasets/terminal-bench-science/terminal-bench-science/0.1.0/leaderboards/v0-1-eval/rows/7cf10b06-3e5b-4723-b98f-c679e773959b",
        "details": [
          {
            "label": "Earth science",
            "value": "0.0% · 24 trials"
          },
          {
            "label": "Engineering science",
            "value": "7.4% · 27 trials"
          },
          {
            "label": "Life science",
            "value": "1.8% · 57 trials"
          },
          {
            "label": "Mathematical science",
            "value": "7.8% · 51 trials"
          },
          {
            "label": "Physical science",
            "value": "0.0% · 51 trials"
          }
        ]
      }
    ]
  }
}