{
  "collectedAt": "2026-09-12",
  "generatedAt": "2026-09-03T22:24:37.984682+00:00",
  "taskCount": 113,
  "unit": "pass@1 is attempt pass rate over scored rollout attempts. pass@4 is tasks with at least one passing rollout divided by tasks attempted. Context-window failures and agent timeouts are scored failures; provider/verifier/network errors are excluded. Efficiency aggregates are over every scored attempt.",
  "boards": [
    {
      "boardId": "deepswe_1_1",
      "name": "DeepSWE v1.1",
      "metric": "Pass@1 %",
      "url": "https://deepswe.datacurve.ai/",
      "snapshotDate": "2026-09-03"
    }
  ],
  "scores": [
    {
      "boardId": "deepswe_1_1",
      "model": "gpt-6-astra",
      "variantLabel": "mini-swe-agent - gpt-6-astra (xhigh)",
      "score": 74.11504424778761,
      "secondary": {
        "agentHarness": "mini-swe-agent",
        "reasoningEffort": "xhigh",
        "ciMinus": 2.865396174075144,
        "ciPlus": 2.865396174075144,
        "ciMethod": "95% run-to-run: SE across repeated whole-benchmark passes (1.96 * std(runs)/sqrt(R))",
        "meanCostUsdPerTask": 6.52377356460177,
        "medianCostPerTaskUsd": 5.6717151,
        "attempted": 452,
        "passed": 335,
        "tasksAttempted": 113,
        "runs": 4,
        "config": "mini_swe_agent_gpt_6_astra_xhigh",
        "costBasis": "Expected launch pricing at all context lengths: $12/M uncached input, $15/M cache writes, $1.20/M cache reads, $50/M output, $2/M compute units."
      },
      "source": "https://deepswe.datacurve.ai/artifacts/v1.1/leaderboard-live.json",
      "checkedAt": "2026-09-12"
    },
    {
      "boardId": "deepswe_1_1",
      "model": "gemini-3.8-flash",
      "variantLabel": "mini-swe-agent - gemini-3-8-flash (high)",
      "score": 73.8255033557047,
      "secondary": {
        "agentHarness": "mini-swe-agent",
        "reasoningEffort": "high",
        "ciMinus": 1.4173132066358676,
        "ciPlus": 1.4173132066358818,
        "ciMethod": "95% run-to-run: SE across repeated whole-benchmark passes (1.96 * std(runs)/sqrt(R))",
        "meanCostUsdPerTask": 2.362349413758389,
        "medianCostPerTaskUsd": 2.1098912999999997,
        "attempted": 447,
        "passed": 330,
        "tasksAttempted": 113,
        "runs": 4,
        "config": "mini_swe_agent_gemini_3_8_flash_high",
        "costBasis": null
      },
      "source": "https://deepswe.datacurve.ai/artifacts/v1.1/leaderboard-live.json",
      "checkedAt": "2026-09-12"
    },
    {
      "boardId": "deepswe_1_1",
      "model": "claude-opus-5",
      "variantLabel": "mini-swe-agent - claude-opus-5 (max)",
      "score": 73.64864864864865,
      "secondary": {
        "agentHarness": "mini-swe-agent",
        "reasoningEffort": "max",
        "ciMinus": 3.872310426371726,
        "ciPlus": 3.8723104263717403,
        "ciMethod": "95% run-to-run: SE across repeated whole-benchmark passes (1.96 * std(runs)/sqrt(R))",
        "meanCostUsdPerTask": 11.837583271396396,
        "medianCostPerTaskUsd": 10.4281505,
        "attempted": 444,
        "passed": 327,
        "tasksAttempted": 113,
        "runs": 4,
        "config": "mini_swe_agent_claude_opus_5_max",
        "costBasis": null
      },
      "source": "https://deepswe.datacurve.ai/artifacts/v1.1/leaderboard-live.json",
      "checkedAt": "2026-09-12"
    },
    {
      "boardId": "deepswe_1_1",
      "model": "gpt-6-astra",
      "variantLabel": "mini-swe-agent - gpt-6-astra (high)",
      "score": 73.23008849557522,
      "secondary": {
        "agentHarness": "mini-swe-agent",
        "reasoningEffort": "high",
        "ciMinus": 3.4234960578730096,
        "ciPlus": 3.4234960578730096,
        "ciMethod": "95% run-to-run: SE across repeated whole-benchmark passes (1.96 * std(runs)/sqrt(R))",
        "meanCostUsdPerTask": 5.723721956194691,
        "medianCostPerTaskUsd": 4.9622669,
        "attempted": 452,
        "passed": 331,
        "tasksAttempted": 113,
        "runs": 4,
        "config": "mini_swe_agent_gpt_6_astra_high",
        "costBasis": "Expected launch pricing at all context lengths: $12/M uncached input, $15/M cache writes, $1.20/M cache reads, $50/M output, $2/M compute units."
      },
      "source": "https://deepswe.datacurve.ai/artifacts/v1.1/leaderboard-live.json",
      "checkedAt": "2026-09-12"
    },
    {
      "boardId": "deepswe_1_1",
      "model": "gpt-6-astra",
      "variantLabel": "mini-swe-agent - gpt-6-astra (max)",
      "score": 73.23008849557522,
      "secondary": {
        "agentHarness": "mini-swe-agent",
        "reasoningEffort": "max",
        "ciMinus": 0.8303197562056539,
        "ciPlus": 0.8303197562056539,
        "ciMethod": "95% run-to-run: SE across repeated whole-benchmark passes (1.96 * std(runs)/sqrt(R))",
        "meanCostUsdPerTask": 12.369032401327434,
        "medianCostPerTaskUsd": 11.036936,
        "attempted": 452,
        "passed": 331,
        "tasksAttempted": 113,
        "runs": 4,
        "config": "mini_swe_agent_gpt_6_astra_max",
        "costBasis": "Expected launch pricing at all context lengths: $12/M uncached input, $15/M cache writes, $1.20/M cache reads, $50/M output, $2/M compute units."
      },
      "source": "https://deepswe.datacurve.ai/artifacts/v1.1/leaderboard-live.json",
      "checkedAt": "2026-09-12"
    },
    {
      "boardId": "deepswe_1_1",
      "model": "claude-opus-5",
      "variantLabel": "mini-swe-agent - claude-opus-5 (xhigh)",
      "score": 73.15436241610739,
      "secondary": {
        "agentHarness": "mini-swe-agent",
        "reasoningEffort": "xhigh",
        "ciMinus": 3.060981821638464,
        "ciPlus": 3.060981821638464,
        "ciMethod": "95% run-to-run: SE across repeated whole-benchmark passes (1.96 * std(runs)/sqrt(R))",
        "meanCostUsdPerTask": 9.07220232326622,
        "medianCostPerTaskUsd": 7.928816750000001,
        "attempted": 447,
        "passed": 327,
        "tasksAttempted": 113,
        "runs": 4,
        "config": "mini_swe_agent_claude_opus_5_xhigh",
        "costBasis": null
      },
      "source": "https://deepswe.datacurve.ai/artifacts/v1.1/leaderboard-live.json",
      "checkedAt": "2026-09-12"
    },
    {
      "boardId": "deepswe_1_1",
      "model": "claude-opus-5",
      "variantLabel": "mini-swe-agent - claude-opus-5 (high)",
      "score": 72.82850779510022,
      "secondary": {
        "agentHarness": "mini-swe-agent",
        "reasoningEffort": "high",
        "ciMinus": 1.9454321571601554,
        "ciPlus": 1.9454321571601554,
        "ciMethod": "95% run-to-run: SE across repeated whole-benchmark passes (1.96 * std(runs)/sqrt(R))",
        "meanCostUsdPerTask": 6.076084489977728,
        "medianCostPerTaskUsd": 4.96882375,
        "attempted": 449,
        "passed": 327,
        "tasksAttempted": 113,
        "runs": 4,
        "config": "mini_swe_agent_claude_opus_5_high",
        "costBasis": null
      },
      "source": "https://deepswe.datacurve.ai/artifacts/v1.1/leaderboard-live.json",
      "checkedAt": "2026-09-12"
    },
    {
      "boardId": "deepswe_1_1",
      "model": "gpt-6-astra",
      "variantLabel": "mini-swe-agent - gpt-6-astra (medium)",
      "score": 72.78761061946902,
      "secondary": {
        "agentHarness": "mini-swe-agent",
        "reasoningEffort": "medium",
        "ciMinus": 2.5896490818318654,
        "ciPlus": 2.5896490818318654,
        "ciMethod": "95% run-to-run: SE across repeated whole-benchmark passes (1.96 * std(runs)/sqrt(R))",
        "meanCostUsdPerTask": 4.379527705752213,
        "medianCostPerTaskUsd": 3.7133415999999997,
        "attempted": 452,
        "passed": 329,
        "tasksAttempted": 113,
        "runs": 4,
        "config": "mini_swe_agent_gpt_6_astra_medium",
        "costBasis": "Expected launch pricing at all context lengths: $12/M uncached input, $15/M cache writes, $1.20/M cache reads, $50/M output, $2/M compute units."
      },
      "source": "https://deepswe.datacurve.ai/artifacts/v1.1/leaderboard-live.json",
      "checkedAt": "2026-09-12"
    },
    {
      "boardId": "deepswe_1_1",
      "model": "gpt-5.6-sol",
      "variantLabel": "mini-swe-agent - gpt-5-6-sol (max)",
      "score": 72.66666666666667,
      "secondary": {
        "agentHarness": "mini-swe-agent",
        "reasoningEffort": "max",
        "ciMinus": 2.8298222498371786,
        "ciPlus": 2.8298222498371786,
        "ciMethod": "95% run-to-run: SE across repeated whole-benchmark passes (1.96 * std(runs)/sqrt(R))",
        "meanCostUsdPerTask": 8.386436346666667,
        "medianCostPerTaskUsd": 6.8414715,
        "attempted": 450,
        "passed": 327,
        "tasksAttempted": 113,
        "runs": 4,
        "config": "mini_swe_agent_gpt_5_6_sol_max",
        "costBasis": null
      },
      "source": "https://deepswe.datacurve.ai/artifacts/v1.1/leaderboard-live.json",
      "checkedAt": "2026-09-12"
    },
    {
      "boardId": "deepswe_1_1",
      "model": "gemini-3.8-flash",
      "variantLabel": "mini-swe-agent - gemini-3-8-flash (medium)",
      "score": 71.01769911504425,
      "secondary": {
        "agentHarness": "mini-swe-agent",
        "reasoningEffort": "medium",
        "ciMinus": 2.2808045728779263,
        "ciPlus": 2.2808045728779263,
        "ciMethod": "95% run-to-run: SE across repeated whole-benchmark passes (1.96 * std(runs)/sqrt(R))",
        "meanCostUsdPerTask": 1.9671024789823006,
        "medianCostPerTaskUsd": 1.7342966249999998,
        "attempted": 452,
        "passed": 321,
        "tasksAttempted": 113,
        "runs": 4,
        "config": "mini_swe_agent_gemini_3_8_flash_medium",
        "costBasis": null
      },
      "source": "https://deepswe.datacurve.ai/artifacts/v1.1/leaderboard-live.json",
      "checkedAt": "2026-09-12"
    },
    {
      "boardId": "deepswe_1_1",
      "model": "gpt-5.6-sol",
      "variantLabel": "mini-swe-agent - gpt-5-6-sol (xhigh)",
      "score": 70.73170731707317,
      "secondary": {
        "agentHarness": "mini-swe-agent",
        "reasoningEffort": "xhigh",
        "ciMinus": 0.8191117217343162,
        "ciPlus": 0.819111721734302,
        "ciMethod": "95% run-to-run: SE across repeated whole-benchmark passes (1.96 * std(runs)/sqrt(R))",
        "meanCostUsdPerTask": 4.703663860310422,
        "medianCostPerTaskUsd": 4.117039,
        "attempted": 451,
        "passed": 319,
        "tasksAttempted": 113,
        "runs": 4,
        "config": "mini_swe_agent_gpt_5_6_sol_xhigh",
        "costBasis": null
      },
      "source": "https://deepswe.datacurve.ai/artifacts/v1.1/leaderboard-live.json",
      "checkedAt": "2026-09-12"
    },
    {
      "boardId": "deepswe_1_1",
      "model": "claude-fable-5",
      "variantLabel": "mini-swe-agent - claude-fable-5 (xhigh)",
      "score": 69.91150442477876,
      "secondary": {
        "agentHarness": "mini-swe-agent",
        "reasoningEffort": "xhigh",
        "ciMinus": 3.2449175754713195,
        "ciPlus": 3.2449175754713195,
        "ciMethod": "95% run-to-run: SE across repeated whole-benchmark passes (1.96 * std(runs)/sqrt(R))",
        "meanCostUsdPerTask": 13.414521495535714,
        "medianCostPerTaskUsd": 11.341082999999998,
        "attempted": 452,
        "passed": 316,
        "tasksAttempted": 113,
        "runs": 4,
        "config": "mini_swe_agent_claude_fable_5_xhigh",
        "costBasis": null
      },
      "source": "https://deepswe.datacurve.ai/artifacts/v1.1/leaderboard-live.json",
      "checkedAt": "2026-09-12"
    },
    {
      "boardId": "deepswe_1_1",
      "model": "claude-fable-5",
      "variantLabel": "mini-swe-agent - claude-fable-5 (max)",
      "score": 69.72477064220183,
      "secondary": {
        "agentHarness": "mini-swe-agent",
        "reasoningEffort": "max",
        "ciMinus": 4.033474250807927,
        "ciPlus": 4.0334742508079415,
        "ciMethod": "95% run-to-run: SE across repeated whole-benchmark passes (1.96 * std(runs)/sqrt(R))",
        "meanCostUsdPerTask": 21.634702092592594,
        "medianCostPerTaskUsd": 19.23457225,
        "attempted": 436,
        "passed": 304,
        "tasksAttempted": 113,
        "runs": 4,
        "config": "mini_swe_agent_claude_fable_5_max",
        "costBasis": null
      },
      "source": "https://deepswe.datacurve.ai/artifacts/v1.1/leaderboard-live.json",
      "checkedAt": "2026-09-12"
    },
    {
      "boardId": "deepswe_1_1",
      "model": "gpt-5.6-terra",
      "variantLabel": "mini-swe-agent - gpt-5-6-terra (max)",
      "score": 69.62305986696231,
      "secondary": {
        "agentHarness": "mini-swe-agent",
        "reasoningEffort": "max",
        "ciMinus": 2.55693895044584,
        "ciPlus": 2.55693895044584,
        "ciMethod": "95% run-to-run: SE across repeated whole-benchmark passes (1.96 * std(runs)/sqrt(R))",
        "meanCostUsdPerTask": 4.945847263858093,
        "medianCostPerTaskUsd": 4.114557,
        "attempted": 451,
        "passed": 314,
        "tasksAttempted": 113,
        "runs": 4,
        "config": "mini_swe_agent_gpt_5_6_terra_max",
        "costBasis": null
      },
      "source": "https://deepswe.datacurve.ai/artifacts/v1.1/leaderboard-live.json",
      "checkedAt": "2026-09-12"
    },
    {
      "boardId": "deepswe_1_1",
      "model": "gpt-5.6-sol",
      "variantLabel": "mini-swe-agent - gpt-5-6-sol (high)",
      "score": 69.40133037694012,
      "secondary": {
        "agentHarness": "mini-swe-agent",
        "reasoningEffort": "high",
        "ciMinus": 1.4321452203772225,
        "ciPlus": 1.4321452203772367,
        "ciMethod": "95% run-to-run: SE across repeated whole-benchmark passes (1.96 * std(runs)/sqrt(R))",
        "meanCostUsdPerTask": 3.4698331108647453,
        "medianCostPerTaskUsd": 2.979421,
        "attempted": 451,
        "passed": 313,
        "tasksAttempted": 113,
        "runs": 4,
        "config": "mini_swe_agent_gpt_5_6_sol_high",
        "costBasis": null
      },
      "source": "https://deepswe.datacurve.ai/artifacts/v1.1/leaderboard-live.json",
      "checkedAt": "2026-09-12"
    },
    {
      "boardId": "deepswe_1_1",
      "model": "glm-5.3",
      "variantLabel": "mini-swe-agent - glm-5-3 (max)",
      "score": 68.95787139689578,
      "secondary": {
        "agentHarness": "mini-swe-agent",
        "reasoningEffort": "max",
        "ciMinus": 3.0186512715586957,
        "ciPlus": 3.01865127155871,
        "ciMethod": "95% run-to-run: SE across repeated whole-benchmark passes (1.96 * std(runs)/sqrt(R))",
        "meanCostUsdPerTask": 3.9933584893126386,
        "medianCostPerTaskUsd": 3.1315945999999992,
        "attempted": 451,
        "passed": 311,
        "tasksAttempted": 113,
        "runs": 4,
        "config": "mini_swe_agent_glm_5_3_max",
        "costBasis": null
      },
      "source": "https://deepswe.datacurve.ai/artifacts/v1.1/leaderboard-live.json",
      "checkedAt": "2026-09-12"
    },
    {
      "boardId": "deepswe_1_1",
      "model": "claude-opus-5",
      "variantLabel": "mini-swe-agent - claude-opus-5 (medium)",
      "score": 68.90380313199105,
      "secondary": {
        "agentHarness": "mini-swe-agent",
        "reasoningEffort": "medium",
        "ciMinus": 1.1725586658931064,
        "ciPlus": 1.1725586658930922,
        "ciMethod": "95% run-to-run: SE across repeated whole-benchmark passes (1.96 * std(runs)/sqrt(R))",
        "meanCostUsdPerTask": 3.28978764541387,
        "medianCostPerTaskUsd": 2.4819885000000004,
        "attempted": 447,
        "passed": 308,
        "tasksAttempted": 113,
        "runs": 4,
        "config": "mini_swe_agent_claude_opus_5_medium",
        "costBasis": null
      },
      "source": "https://deepswe.datacurve.ai/artifacts/v1.1/leaderboard-live.json",
      "checkedAt": "2026-09-12"
    },
    {
      "boardId": "deepswe_1_1",
      "model": "claude-fable-5",
      "variantLabel": "mini-swe-agent - claude-fable-5 (high)",
      "score": 68.6046511627907,
      "secondary": {
        "agentHarness": "mini-swe-agent",
        "reasoningEffort": "high",
        "ciMinus": 1.120842717430392,
        "ciPlus": 1.120842717430392,
        "ciMethod": "95% run-to-run: SE across repeated whole-benchmark passes (1.96 * std(runs)/sqrt(R))",
        "meanCostUsdPerTask": 9.177638214788733,
        "medianCostPerTaskUsd": 7.322686000000002,
        "attempted": 430,
        "passed": 295,
        "tasksAttempted": 113,
        "runs": 4,
        "config": "mini_swe_agent_claude_fable_5_high",
        "costBasis": null
      },
      "source": "https://deepswe.datacurve.ai/artifacts/v1.1/leaderboard-live.json",
      "checkedAt": "2026-09-12"
    },
    {
      "boardId": "deepswe_1_1",
      "model": "kimi-k3",
      "variantLabel": "mini-swe-agent - kimi-k3 (max)",
      "score": 68.51441241685144,
      "secondary": {
        "agentHarness": "mini-swe-agent",
        "reasoningEffort": "max",
        "ciMinus": 4.537014079282315,
        "ciPlus": 4.537014079282315,
        "ciMethod": "95% run-to-run: SE across repeated whole-benchmark passes (1.96 * std(runs)/sqrt(R))",
        "meanCostUsdPerTask": 4.654682129933482,
        "medianCostPerTaskUsd": 3.6024987,
        "attempted": 451,
        "passed": 309,
        "tasksAttempted": 113,
        "runs": 4,
        "config": "mini_swe_agent_kimi_k3_max",
        "costBasis": null
      },
      "source": "https://deepswe.datacurve.ai/artifacts/v1.1/leaderboard-live.json",
      "checkedAt": "2026-09-12"
    },
    {
      "boardId": "deepswe_1_1",
      "model": "grok-4.6",
      "variantLabel": "mini-swe-agent - grok-4-6 (medium)",
      "score": 67.47787610619469,
      "secondary": {
        "agentHarness": "mini-swe-agent",
        "reasoningEffort": "medium",
        "ciMinus": 2.280804572877912,
        "ciPlus": 2.280804572877912,
        "ciMethod": "95% run-to-run: SE across repeated whole-benchmark passes (1.96 * std(runs)/sqrt(R))",
        "meanCostUsdPerTask": 3.4489837522123894,
        "medianCostPerTaskUsd": 2.945834,
        "attempted": 452,
        "passed": 305,
        "tasksAttempted": 113,
        "runs": 4,
        "config": "mini_swe_agent_grok_4_6_medium",
        "costBasis": null
      },
      "source": "https://deepswe.datacurve.ai/artifacts/v1.1/leaderboard-live.json",
      "checkedAt": "2026-09-12"
    },
    {
      "boardId": "deepswe_1_1",
      "model": "gpt-5.6-luna",
      "variantLabel": "mini-swe-agent - gpt-5-6-luna (max)",
      "score": 67.1875,
      "secondary": {
        "agentHarness": "mini-swe-agent",
        "reasoningEffort": "max",
        "ciMinus": 3.9920012356710473,
        "ciPlus": 3.9920012356710544,
        "ciMethod": "95% run-to-run: SE across repeated whole-benchmark passes (1.96 * std(runs)/sqrt(R))",
        "meanCostUsdPerTask": 3.028116810267857,
        "medianCostPerTaskUsd": 2.2920917999999997,
        "attempted": 448,
        "passed": 301,
        "tasksAttempted": 113,
        "runs": 4,
        "config": "mini_swe_agent_gpt_5_6_luna_max",
        "costBasis": null
      },
      "source": "https://deepswe.datacurve.ai/artifacts/v1.1/leaderboard-live.json",
      "checkedAt": "2026-09-12"
    },
    {
      "boardId": "deepswe_1_1",
      "model": "gpt-5.5",
      "variantLabel": "mini-swe-agent - gpt-5-5 (xhigh)",
      "score": 67.03539823008849,
      "secondary": {
        "agentHarness": "mini-swe-agent",
        "reasoningEffort": "xhigh",
        "ciMinus": 6.4656463409088545,
        "ciPlus": 6.465646340908862,
        "ciMethod": "95% run-to-run: SE across repeated whole-benchmark passes (1.96 * std(runs)/sqrt(R))",
        "meanCostUsdPerTask": 7.226236674778761,
        "medianCostPerTaskUsd": 6.110482499999999,
        "attempted": 452,
        "passed": 303,
        "tasksAttempted": 113,
        "runs": 4,
        "config": "mini_swe_agent_gpt_5_5_xhigh",
        "costBasis": null
      },
      "source": "https://deepswe.datacurve.ai/artifacts/v1.1/leaderboard-live.json",
      "checkedAt": "2026-09-12"
    },
    {
      "boardId": "deepswe_1_1",
      "model": "gpt-6-astra",
      "variantLabel": "mini-swe-agent - gpt-6-astra (low)",
      "score": 67.03539823008849,
      "secondary": {
        "agentHarness": "mini-swe-agent",
        "reasoningEffort": "low",
        "ciMinus": 1.300861051685871,
        "ciPlus": 1.3008610516858852,
        "ciMethod": "95% run-to-run: SE across repeated whole-benchmark passes (1.96 * std(runs)/sqrt(R))",
        "meanCostUsdPerTask": 2.188762706637168,
        "medianCostPerTaskUsd": 1.7209886,
        "attempted": 452,
        "passed": 303,
        "tasksAttempted": 113,
        "runs": 4,
        "config": "mini_swe_agent_gpt_6_astra_low",
        "costBasis": "Expected launch pricing at all context lengths: $12/M uncached input, $15/M cache writes, $1.20/M cache reads, $50/M output, $2/M compute units."
      },
      "source": "https://deepswe.datacurve.ai/artifacts/v1.1/leaderboard-live.json",
      "checkedAt": "2026-09-12"
    },
    {
      "boardId": "deepswe_1_1",
      "model": "grok-4.6",
      "variantLabel": "mini-swe-agent - grok-4-6 (xhigh)",
      "score": 66.74057649667405,
      "secondary": {
        "agentHarness": "mini-swe-agent",
        "reasoningEffort": "xhigh",
        "ciMinus": 2.175684224405046,
        "ciPlus": 2.175684224405046,
        "ciMethod": "95% run-to-run: SE across repeated whole-benchmark passes (1.96 * std(runs)/sqrt(R))",
        "meanCostUsdPerTask": 5.497668745011087,
        "medianCostPerTaskUsd": 4.63937,
        "attempted": 451,
        "passed": 301,
        "tasksAttempted": 113,
        "runs": 4,
        "config": "mini_swe_agent_grok_4_6_xhigh",
        "costBasis": null
      },
      "source": "https://deepswe.datacurve.ai/artifacts/v1.1/leaderboard-live.json",
      "checkedAt": "2026-09-12"
    },
    {
      "boardId": "deepswe_1_1",
      "model": "gemini-3.7-flash",
      "variantLabel": "mini-swe-agent - gemini-3-7-flash (medium)",
      "score": 65.48672566371681,
      "secondary": {
        "agentHarness": "mini-swe-agent",
        "reasoningEffort": "medium",
        "ciMinus": 3.086532276415518,
        "ciPlus": 3.086532276415525,
        "ciMethod": "95% run-to-run: SE across repeated whole-benchmark passes (1.96 * std(runs)/sqrt(R))",
        "meanCostUsdPerTask": 2.0251351826880533,
        "medianCostPerTaskUsd": 1.65422055,
        "attempted": 452,
        "passed": 296,
        "tasksAttempted": 113,
        "runs": 4,
        "config": "mini_swe_agent_gemini_3_7_flash_medium",
        "costBasis": null
      },
      "source": "https://deepswe.datacurve.ai/artifacts/v1.1/leaderboard-live.json",
      "checkedAt": "2026-09-12"
    },
    {
      "boardId": "deepswe_1_1",
      "model": "claude-fable-5",
      "variantLabel": "mini-swe-agent - claude-fable-5 (medium)",
      "score": 65.36697247706422,
      "secondary": {
        "agentHarness": "mini-swe-agent",
        "reasoningEffort": "medium",
        "ciMinus": 4.421928879238166,
        "ciPlus": 4.421928879238166,
        "ciMethod": "95% run-to-run: SE across repeated whole-benchmark passes (1.96 * std(runs)/sqrt(R))",
        "meanCostUsdPerTask": 6.088186528935186,
        "medianCostPerTaskUsd": 4.62221125,
        "attempted": 436,
        "passed": 285,
        "tasksAttempted": 113,
        "runs": 4,
        "config": "mini_swe_agent_claude_fable_5_medium",
        "costBasis": null
      },
      "source": "https://deepswe.datacurve.ai/artifacts/v1.1/leaderboard-live.json",
      "checkedAt": "2026-09-12"
    },
    {
      "boardId": "deepswe_1_1",
      "model": "gemini-3.7-flash",
      "variantLabel": "mini-swe-agent - gemini-3-7-flash (high)",
      "score": 65.2654867256637,
      "secondary": {
        "agentHarness": "mini-swe-agent",
        "reasoningEffort": "high",
        "ciMinus": 1.7878625067843146,
        "ciPlus": 1.7878625067843217,
        "ciMethod": "95% run-to-run: SE across repeated whole-benchmark passes (1.96 * std(runs)/sqrt(R))",
        "meanCostUsdPerTask": 2.1762860351216813,
        "medianCostPerTaskUsd": 1.9044440625,
        "attempted": 452,
        "passed": 295,
        "tasksAttempted": 113,
        "runs": 4,
        "config": "mini_swe_agent_gemini_3_7_flash_high",
        "costBasis": null
      },
      "source": "https://deepswe.datacurve.ai/artifacts/v1.1/leaderboard-live.json",
      "checkedAt": "2026-09-12"
    },
    {
      "boardId": "deepswe_1_1",
      "model": "grok-4.6",
      "variantLabel": "mini-swe-agent - grok-4-6 (high)",
      "score": 65.18847006651885,
      "secondary": {
        "agentHarness": "mini-swe-agent",
        "reasoningEffort": "high",
        "ciMinus": 1.5336008186572982,
        "ciPlus": 1.5336008186572911,
        "ciMethod": "95% run-to-run: SE across repeated whole-benchmark passes (1.96 * std(runs)/sqrt(R))",
        "meanCostUsdPerTask": 4.3848732239467845,
        "medianCostPerTaskUsd": 3.693604,
        "attempted": 451,
        "passed": 294,
        "tasksAttempted": 113,
        "runs": 4,
        "config": "mini_swe_agent_grok_4_6_high",
        "costBasis": null
      },
      "source": "https://deepswe.datacurve.ai/artifacts/v1.1/leaderboard-live.json",
      "checkedAt": "2026-09-12"
    },
    {
      "boardId": "deepswe_1_1",
      "model": "gpt-5.5",
      "variantLabel": "mini-swe-agent - gpt-5-5 (high)",
      "score": 64.38053097345133,
      "secondary": {
        "agentHarness": "mini-swe-agent",
        "reasoningEffort": "high",
        "ciMinus": 3.116842649505479,
        "ciPlus": 3.1168426495054575,
        "ciMethod": "95% run-to-run: SE across repeated whole-benchmark passes (1.96 * std(runs)/sqrt(R))",
        "meanCostUsdPerTask": 5.100437004424779,
        "medianCostPerTaskUsd": 4.582879500000001,
        "attempted": 452,
        "passed": 291,
        "tasksAttempted": 113,
        "runs": 4,
        "config": "mini_swe_agent_gpt_5_5_high",
        "costBasis": null
      },
      "source": "https://deepswe.datacurve.ai/artifacts/v1.1/leaderboard-live.json",
      "checkedAt": "2026-09-12"
    },
    {
      "boardId": "deepswe_1_1",
      "model": "glm-5.3-flash",
      "variantLabel": "mini-swe-agent - glm-5-3-flash (max)",
      "score": 63.39285714285714,
      "secondary": {
        "agentHarness": "mini-swe-agent",
        "reasoningEffort": "max",
        "ciMinus": 4.377970751693638,
        "ciPlus": 4.377970751693638,
        "ciMethod": "95% run-to-run: SE across repeated whole-benchmark passes (1.96 * std(runs)/sqrt(R))",
        "meanCostUsdPerTask": 0.48196371245535713,
        "medianCostPerTaskUsd": 0.389975125,
        "attempted": 448,
        "passed": 284,
        "tasksAttempted": 113,
        "runs": 4,
        "config": "mini_swe_agent_glm_5_3_flash_max",
        "costBasis": null
      },
      "source": "https://deepswe.datacurve.ai/artifacts/v1.1/leaderboard-live.json",
      "checkedAt": "2026-09-12"
    },
    {
      "boardId": "deepswe_1_1",
      "model": "deepseek-v4-pro",
      "variantLabel": "mini-swe-agent - deepseek-v4-pro (max)",
      "score": 62.83185840707964,
      "secondary": {
        "agentHarness": "mini-swe-agent",
        "reasoningEffort": "max",
        "ciMinus": 6.333430597228876,
        "ciPlus": 6.333430597228876,
        "ciMethod": "95% run-to-run: SE across repeated whole-benchmark passes (1.96 * std(runs)/sqrt(R))",
        "meanCostUsdPerTask": 0.24138687892256638,
        "medianCostPerTaskUsd": 0.2232531505000001,
        "attempted": 452,
        "passed": 284,
        "tasksAttempted": 113,
        "runs": 4,
        "config": "mini_swe_agent_deepseek_v4_pro_max",
        "costBasis": null
      },
      "source": "https://deepswe.datacurve.ai/artifacts/v1.1/leaderboard-live.json",
      "checkedAt": "2026-09-12"
    },
    {
      "boardId": "deepswe_1_1",
      "model": "gpt-5.6-sol",
      "variantLabel": "mini-swe-agent - gpt-5-6-sol (medium)",
      "score": 61.06194690265486,
      "secondary": {
        "agentHarness": "mini-swe-agent",
        "reasoningEffort": "medium",
        "ciMinus": 1.5833576493072101,
        "ciPlus": 1.5833576493072172,
        "ciMethod": "95% run-to-run: SE across repeated whole-benchmark passes (1.96 * std(runs)/sqrt(R))",
        "meanCostUsdPerTask": 1.8620318075221238,
        "medianCostPerTaskUsd": 1.545414,
        "attempted": 452,
        "passed": 276,
        "tasksAttempted": 113,
        "runs": 4,
        "config": "mini_swe_agent_gpt_5_6_sol_medium",
        "costBasis": null
      },
      "source": "https://deepswe.datacurve.ai/artifacts/v1.1/leaderboard-live.json",
      "checkedAt": "2026-09-12"
    },
    {
      "boardId": "deepswe_1_1",
      "model": "gpt-5.6-terra",
      "variantLabel": "mini-swe-agent - gpt-5-6-terra (xhigh)",
      "score": 60.17699115044248,
      "secondary": {
        "agentHarness": "mini-swe-agent",
        "reasoningEffort": "xhigh",
        "ciMinus": 2.124297201927128,
        "ciPlus": 2.124297201927128,
        "ciMethod": "95% run-to-run: SE across repeated whole-benchmark passes (1.96 * std(runs)/sqrt(R))",
        "meanCostUsdPerTask": 2.127191725663717,
        "medianCostPerTaskUsd": 1.8670102499999999,
        "attempted": 452,
        "passed": 272,
        "tasksAttempted": 113,
        "runs": 4,
        "config": "mini_swe_agent_gpt_5_6_terra_xhigh",
        "costBasis": null
      },
      "source": "https://deepswe.datacurve.ai/artifacts/v1.1/leaderboard-live.json",
      "checkedAt": "2026-09-12"
    },
    {
      "boardId": "deepswe_1_1",
      "model": "claude-fable-5",
      "variantLabel": "mini-swe-agent - claude-fable-5 (low)",
      "score": 59.58429561200924,
      "secondary": {
        "agentHarness": "mini-swe-agent",
        "reasoningEffort": "low",
        "ciMinus": 2.7940054451297343,
        "ciPlus": 2.7940054451297343,
        "ciMethod": "95% run-to-run: SE across repeated whole-benchmark passes (1.96 * std(runs)/sqrt(R))",
        "meanCostUsdPerTask": 3.757873125874126,
        "medianCostPerTaskUsd": 2.7559204999999998,
        "attempted": 433,
        "passed": 258,
        "tasksAttempted": 113,
        "runs": 4,
        "config": "mini_swe_agent_claude_fable_5_low",
        "costBasis": null
      },
      "source": "https://deepswe.datacurve.ai/artifacts/v1.1/leaderboard-live.json",
      "checkedAt": "2026-09-12"
    },
    {
      "boardId": "deepswe_1_1",
      "model": "claude-opus-4.8",
      "variantLabel": "mini-swe-agent - claude-opus-4-8 (max)",
      "score": 58.97435897435898,
      "secondary": {
        "agentHarness": "mini-swe-agent",
        "reasoningEffort": "max",
        "ciMinus": 1.7648153474292485,
        "ciPlus": 1.7648153474292485,
        "ciMethod": "95% run-to-run: SE across repeated whole-benchmark passes (1.96 * std(runs)/sqrt(R))",
        "meanCostUsdPerTask": 13.222583593240094,
        "medianCostPerTaskUsd": 12.353505999999996,
        "attempted": 429,
        "passed": 253,
        "tasksAttempted": 111,
        "runs": 4,
        "config": "mini_swe_agent_claude_opus_4_8_max",
        "costBasis": null
      },
      "source": "https://deepswe.datacurve.ai/artifacts/v1.1/leaderboard-live.json",
      "checkedAt": "2026-09-12"
    },
    {
      "boardId": "deepswe_1_1",
      "model": "claude-opus-5",
      "variantLabel": "mini-swe-agent - claude-opus-5 (low)",
      "score": 58.12917594654788,
      "secondary": {
        "agentHarness": "mini-swe-agent",
        "reasoningEffort": "low",
        "ciMinus": 2.3306879466842716,
        "ciPlus": 2.3306879466842787,
        "ciMethod": "95% run-to-run: SE across repeated whole-benchmark passes (1.96 * std(runs)/sqrt(R))",
        "meanCostUsdPerTask": 1.6626223919821828,
        "medianCostPerTaskUsd": 1.2162672499999996,
        "attempted": 449,
        "passed": 261,
        "tasksAttempted": 113,
        "runs": 4,
        "config": "mini_swe_agent_claude_opus_5_low",
        "costBasis": null
      },
      "source": "https://deepswe.datacurve.ai/artifacts/v1.1/leaderboard-live.json",
      "checkedAt": "2026-09-12"
    },
    {
      "boardId": "deepswe_1_1",
      "model": "qwen3.8-max",
      "variantLabel": "mini-swe-agent - qwen3-8-max (xhigh)",
      "score": 57.46102449888642,
      "secondary": {
        "agentHarness": "mini-swe-agent",
        "reasoningEffort": "xhigh",
        "ciMinus": 2.6627926725075426,
        "ciPlus": 2.6627926725075426,
        "ciMethod": "95% run-to-run: SE across repeated whole-benchmark passes (1.96 * std(runs)/sqrt(R))",
        "meanCostUsdPerTask": 3.7290646500445437,
        "medianCostPerTaskUsd": 3.132689060000001,
        "attempted": 449,
        "passed": 258,
        "tasksAttempted": 113,
        "runs": 4,
        "config": "mini_swe_agent_qwen3_8_max_xhigh",
        "costBasis": null
      },
      "source": "https://deepswe.datacurve.ai/artifacts/v1.1/leaderboard-live.json",
      "checkedAt": "2026-09-12"
    },
    {
      "boardId": "deepswe_1_1",
      "model": "gpt-5.6-luna",
      "variantLabel": "mini-swe-agent - gpt-5-6-luna (xhigh)",
      "score": 56.85840707964602,
      "secondary": {
        "agentHarness": "mini-swe-agent",
        "reasoningEffort": "xhigh",
        "ciMinus": 2.1681017528097897,
        "ciPlus": 2.168101752809797,
        "ciMethod": "95% run-to-run: SE across repeated whole-benchmark passes (1.96 * std(runs)/sqrt(R))",
        "meanCostUsdPerTask": 1.5356192685840708,
        "medianCostPerTaskUsd": 1.2449332,
        "attempted": 452,
        "passed": 257,
        "tasksAttempted": 113,
        "runs": 4,
        "config": "mini_swe_agent_gpt_5_6_luna_xhigh",
        "costBasis": null
      },
      "source": "https://deepswe.datacurve.ai/artifacts/v1.1/leaderboard-live.json",
      "checkedAt": "2026-09-12"
    },
    {
      "boardId": "deepswe_1_1",
      "model": "muse-spark-1.2",
      "variantLabel": "mini-swe-agent - muse-spark-1-2 (xhigh)",
      "score": 54.86725663716814,
      "secondary": {
        "agentHarness": "mini-swe-agent",
        "reasoningEffort": "xhigh",
        "ciMinus": 2.124297201927128,
        "ciPlus": 2.124297201927128,
        "ciMethod": "95% run-to-run: SE across repeated whole-benchmark passes (1.96 * std(runs)/sqrt(R))",
        "meanCostUsdPerTask": 3.6955048180309733,
        "medianCostPerTaskUsd": 2.6840506,
        "attempted": 452,
        "passed": 248,
        "tasksAttempted": 113,
        "runs": 4,
        "config": "mini_swe_agent_muse_spark_1_2_xhigh",
        "costBasis": null
      },
      "source": "https://deepswe.datacurve.ai/artifacts/v1.1/leaderboard-live.json",
      "checkedAt": "2026-09-12"
    },
    {
      "boardId": "deepswe_1_1",
      "model": "claude-opus-4.8",
      "variantLabel": "mini-swe-agent - claude-opus-4-8 (xhigh)",
      "score": 54.36241610738255,
      "secondary": {
        "agentHarness": "mini-swe-agent",
        "reasoningEffort": "xhigh",
        "ciMinus": 3.7134306280944713,
        "ciPlus": 3.7134306280944713,
        "ciMethod": "95% run-to-run: SE across repeated whole-benchmark passes (1.96 * std(runs)/sqrt(R))",
        "meanCostUsdPerTask": 8.006355704138702,
        "medianCostPerTaskUsd": 7.049357750000001,
        "attempted": 447,
        "passed": 243,
        "tasksAttempted": 113,
        "runs": 4,
        "config": "mini_swe_agent_claude_opus_4_8_xhigh",
        "costBasis": null
      },
      "source": "https://deepswe.datacurve.ai/artifacts/v1.1/leaderboard-live.json",
      "checkedAt": "2026-09-12"
    },
    {
      "boardId": "deepswe_1_1",
      "model": "gpt-5.5",
      "variantLabel": "mini-swe-agent - gpt-5-5 (medium)",
      "score": 53.98230088495575,
      "secondary": {
        "agentHarness": "mini-swe-agent",
        "reasoningEffort": "medium",
        "ciMinus": 2.5530874952909883,
        "ciPlus": 2.5530874952909812,
        "ciMethod": "95% run-to-run: SE across repeated whole-benchmark passes (1.96 * std(runs)/sqrt(R))",
        "meanCostUsdPerTask": 2.749342734513274,
        "medianCostPerTaskUsd": 2.436942,
        "attempted": 452,
        "passed": 244,
        "tasksAttempted": 113,
        "runs": 4,
        "config": "mini_swe_agent_gpt_5_5_medium",
        "costBasis": null
      },
      "source": "https://deepswe.datacurve.ai/artifacts/v1.1/leaderboard-live.json",
      "checkedAt": "2026-09-12"
    },
    {
      "boardId": "deepswe_1_1",
      "model": "claude-sonnet-5",
      "variantLabel": "mini-swe-agent - claude-sonnet-5 (max)",
      "score": 53.84615384615385,
      "secondary": {
        "agentHarness": "mini-swe-agent",
        "reasoningEffort": "max",
        "ciMinus": 4.2369164701743856,
        "ciPlus": 4.2369164701743856,
        "ciMethod": "95% run-to-run: SE across repeated whole-benchmark passes (1.96 * std(runs)/sqrt(R))",
        "meanCostUsdPerTask": 26.399858950791852,
        "medianCostPerTaskUsd": 23.277376725000003,
        "attempted": 442,
        "passed": 238,
        "tasksAttempted": 113,
        "runs": 4,
        "config": "mini_swe_agent_claude_sonnet_5_max",
        "costBasis": null
      },
      "source": "https://deepswe.datacurve.ai/artifacts/v1.1/leaderboard-live.json",
      "checkedAt": "2026-09-12"
    },
    {
      "boardId": "deepswe_1_1",
      "model": "gemini-3.7-flash",
      "variantLabel": "mini-swe-agent - gemini-3-7-flash (low)",
      "score": 53.76106194690266,
      "secondary": {
        "agentHarness": "mini-swe-agent",
        "reasoningEffort": "low",
        "ciMinus": 2.5896490818318725,
        "ciPlus": 2.5896490818318654,
        "ciMethod": "95% run-to-run: SE across repeated whole-benchmark passes (1.96 * std(runs)/sqrt(R))",
        "meanCostUsdPerTask": 1.8322645629424779,
        "medianCostPerTaskUsd": 1.3913454375,
        "attempted": 452,
        "passed": 243,
        "tasksAttempted": 113,
        "runs": 4,
        "config": "mini_swe_agent_gemini_3_7_flash_low",
        "costBasis": null
      },
      "source": "https://deepswe.datacurve.ai/artifacts/v1.1/leaderboard-live.json",
      "checkedAt": "2026-09-12"
    },
    {
      "boardId": "deepswe_1_1",
      "model": "gpt-5.6-terra",
      "variantLabel": "mini-swe-agent - gpt-5-6-terra (high)",
      "score": 53.76106194690266,
      "secondary": {
        "agentHarness": "mini-swe-agent",
        "reasoningEffort": "high",
        "ciMinus": 4.328970467213551,
        "ciPlus": 4.328970467213551,
        "ciMethod": "95% run-to-run: SE across repeated whole-benchmark passes (1.96 * std(runs)/sqrt(R))",
        "meanCostUsdPerTask": 1.1343762765486725,
        "medianCostPerTaskUsd": 1.031709,
        "attempted": 452,
        "passed": 243,
        "tasksAttempted": 113,
        "runs": 4,
        "config": "mini_swe_agent_gpt_5_6_terra_high",
        "costBasis": null
      },
      "source": "https://deepswe.datacurve.ai/artifacts/v1.1/leaderboard-live.json",
      "checkedAt": "2026-09-12"
    },
    {
      "boardId": "deepswe_1_1",
      "model": "grok-4.5",
      "variantLabel": "mini-swe-agent - grok-4-5 (high)",
      "score": 53.76106194690266,
      "secondary": {
        "agentHarness": "mini-swe-agent",
        "reasoningEffort": "high",
        "ciMinus": 2.280804572877919,
        "ciPlus": 2.280804572877912,
        "ciMethod": "95% run-to-run: SE across repeated whole-benchmark passes (1.96 * std(runs)/sqrt(R))",
        "meanCostUsdPerTask": 2.4157481725663716,
        "medianCostPerTaskUsd": 2.02311,
        "attempted": 452,
        "passed": 243,
        "tasksAttempted": 113,
        "runs": 4,
        "config": "mini_swe_agent_grok_4_5_high",
        "costBasis": null
      },
      "source": "https://deepswe.datacurve.ai/artifacts/v1.1/leaderboard-live.json",
      "checkedAt": "2026-09-12"
    },
    {
      "boardId": "deepswe_1_1",
      "model": "deepseek-v4-flash",
      "variantLabel": "mini-swe-agent - deepseek-v4-flash (max)",
      "score": 53.31858407079646,
      "secondary": {
        "agentHarness": "mini-swe-agent",
        "reasoningEffort": "max",
        "ciMinus": 3.5669502150324277,
        "ciPlus": 3.5669502150324277,
        "ciMethod": "95% run-to-run: SE across repeated whole-benchmark passes (1.96 * std(runs)/sqrt(R))",
        "meanCostUsdPerTask": 0.10023560370265487,
        "medianCostPerTaskUsd": 0.09124898719999996,
        "attempted": 452,
        "passed": 241,
        "tasksAttempted": 113,
        "runs": 4,
        "config": "mini_swe_agent_deepseek_v4_flash_max",
        "costBasis": null
      },
      "source": "https://deepswe.datacurve.ai/artifacts/v1.1/leaderboard-live.json",
      "checkedAt": "2026-09-12"
    },
    {
      "boardId": "deepswe_1_1",
      "model": "muse-spark-1.1",
      "variantLabel": "mini-swe-agent - muse-spark-1-1 (xhigh)",
      "score": 53.31858407079646,
      "secondary": {
        "agentHarness": "mini-swe-agent",
        "reasoningEffort": "xhigh",
        "ciMinus": 3.0353424539337084,
        "ciPlus": 3.0353424539337084,
        "ciMethod": "95% run-to-run: SE across repeated whole-benchmark passes (1.96 * std(runs)/sqrt(R))",
        "meanCostUsdPerTask": 2.361142644469026,
        "medianCostPerTaskUsd": 1.7171158,
        "attempted": 452,
        "passed": 241,
        "tasksAttempted": 113,
        "runs": 4,
        "config": "mini_swe_agent_muse_spark_1_1_xhigh",
        "costBasis": null
      },
      "source": "https://deepswe.datacurve.ai/artifacts/v1.1/leaderboard-live.json",
      "checkedAt": "2026-09-12"
    },
    {
      "boardId": "deepswe_1_1",
      "model": "claude-opus-4.8",
      "variantLabel": "mini-swe-agent - claude-opus-4-8 (high)",
      "score": 51.76991150442478,
      "secondary": {
        "agentHarness": "mini-swe-agent",
        "reasoningEffort": "high",
        "ciMinus": 4.561609145755845,
        "ciPlus": 4.561609145755838,
        "ciMethod": "95% run-to-run: SE across repeated whole-benchmark passes (1.96 * std(runs)/sqrt(R))",
        "meanCostUsdPerTask": 4.281940017699115,
        "medianCostPerTaskUsd": 3.6635345,
        "attempted": 452,
        "passed": 234,
        "tasksAttempted": 113,
        "runs": 4,
        "config": "mini_swe_agent_claude_opus_4_8_high",
        "costBasis": null
      },
      "source": "https://deepswe.datacurve.ai/artifacts/v1.1/leaderboard-live.json",
      "checkedAt": "2026-09-12"
    },
    {
      "boardId": "deepswe_1_1",
      "model": "gpt-5.4",
      "variantLabel": "mini-swe-agent - gpt-5-4 (xhigh)",
      "score": 51.76991150442478,
      "secondary": {
        "agentHarness": "mini-swe-agent",
        "reasoningEffort": "xhigh",
        "ciMinus": 1.502104956738286,
        "ciPlus": 1.502104956738279,
        "ciMethod": "95% run-to-run: SE across repeated whole-benchmark passes (1.96 * std(runs)/sqrt(R))",
        "meanCostUsdPerTask": 5.6524577245575225,
        "medianCostPerTaskUsd": 4.3528565,
        "attempted": 452,
        "passed": 234,
        "tasksAttempted": 113,
        "runs": 4,
        "config": "mini_swe_agent_gpt_5_4_xhigh",
        "costBasis": null
      },
      "source": "https://deepswe.datacurve.ai/artifacts/v1.1/leaderboard-live.json",
      "checkedAt": "2026-09-12"
    },
    {
      "boardId": "deepswe_1_1",
      "model": "claude-sonnet-5",
      "variantLabel": "mini-swe-agent - claude-sonnet-5 (xhigh)",
      "score": 49.667405764966745,
      "secondary": {
        "agentHarness": "mini-swe-agent",
        "reasoningEffort": "xhigh",
        "ciMinus": 3.4545489916778678,
        "ciPlus": 3.4545489916778607,
        "ciMethod": "95% run-to-run: SE across repeated whole-benchmark passes (1.96 * std(runs)/sqrt(R))",
        "meanCostUsdPerTask": 11.890634540022173,
        "medianCostPerTaskUsd": 9.965361750000001,
        "attempted": 451,
        "passed": 224,
        "tasksAttempted": 113,
        "runs": 4,
        "config": "mini_swe_agent_claude_sonnet_5_xhigh",
        "costBasis": null
      },
      "source": "https://deepswe.datacurve.ai/artifacts/v1.1/leaderboard-live.json",
      "checkedAt": "2026-09-12"
    },
    {
      "boardId": "deepswe_1_1",
      "model": "claude-opus-4.8",
      "variantLabel": "mini-swe-agent - claude-opus-4-8 (medium)",
      "score": 48.67256637168141,
      "secondary": {
        "agentHarness": "mini-swe-agent",
        "reasoningEffort": "medium",
        "ciMinus": 2.239205861737453,
        "ciPlus": 2.239205861737446,
        "ciMethod": "95% run-to-run: SE across repeated whole-benchmark passes (1.96 * std(runs)/sqrt(R))",
        "meanCostUsdPerTask": 3.4438571028761062,
        "medianCostPerTaskUsd": 2.84043425,
        "attempted": 452,
        "passed": 220,
        "tasksAttempted": 113,
        "runs": 4,
        "config": "mini_swe_agent_claude_opus_4_8_medium",
        "costBasis": null
      },
      "source": "https://deepswe.datacurve.ai/artifacts/v1.1/leaderboard-live.json",
      "checkedAt": "2026-09-12"
    },
    {
      "boardId": "deepswe_1_1",
      "model": "claude-sonnet-5",
      "variantLabel": "mini-swe-agent - claude-sonnet-5 (high)",
      "score": 48.23008849557522,
      "secondary": {
        "agentHarness": "mini-swe-agent",
        "reasoningEffort": "high",
        "ciMinus": 4.506314870214837,
        "ciPlus": 4.506314870214844,
        "ciMethod": "95% run-to-run: SE across repeated whole-benchmark passes (1.96 * std(runs)/sqrt(R))",
        "meanCostUsdPerTask": 7.425560161197339,
        "medianCostPerTaskUsd": 5.833649100000001,
        "attempted": 452,
        "passed": 218,
        "tasksAttempted": 113,
        "runs": 4,
        "config": "mini_swe_agent_claude_sonnet_5_high",
        "costBasis": null
      },
      "source": "https://deepswe.datacurve.ai/artifacts/v1.1/leaderboard-live.json",
      "checkedAt": "2026-09-12"
    },
    {
      "boardId": "deepswe_1_1",
      "model": "gemini-3.6-flash",
      "variantLabel": "mini-swe-agent - gemini-3-6-flash (high)",
      "score": 46.68141592920354,
      "secondary": {
        "agentHarness": "mini-swe-agent",
        "reasoningEffort": "high",
        "ciMinus": 3.7048538992472757,
        "ciPlus": 3.7048538992472757,
        "ciMethod": "95% run-to-run: SE across repeated whole-benchmark passes (1.96 * std(runs)/sqrt(R))",
        "meanCostUsdPerTask": 4.419013277333334,
        "medianCostPerTaskUsd": 3.444695325000001,
        "attempted": 452,
        "passed": 211,
        "tasksAttempted": 113,
        "runs": 4,
        "config": "mini_swe_agent_gemini_3_6_flash_high",
        "costBasis": null
      },
      "source": "https://deepswe.datacurve.ai/artifacts/v1.1/leaderboard-live.json",
      "checkedAt": "2026-09-12"
    },
    {
      "boardId": "deepswe_1_1",
      "model": "gpt-5.6-sol",
      "variantLabel": "mini-swe-agent - gpt-5-6-sol (low)",
      "score": 45.35398230088495,
      "secondary": {
        "agentHarness": "mini-swe-agent",
        "reasoningEffort": "low",
        "ciMinus": 2.388194671458919,
        "ciPlus": 2.388194671458926,
        "ciMethod": "95% run-to-run: SE across repeated whole-benchmark passes (1.96 * std(runs)/sqrt(R))",
        "meanCostUsdPerTask": 1.0742664646017699,
        "medianCostPerTaskUsd": 0.944557,
        "attempted": 452,
        "passed": 205,
        "tasksAttempted": 113,
        "runs": 4,
        "config": "mini_swe_agent_gpt_5_6_sol_low",
        "costBasis": null
      },
      "source": "https://deepswe.datacurve.ai/artifacts/v1.1/leaderboard-live.json",
      "checkedAt": "2026-09-12"
    },
    {
      "boardId": "deepswe_1_1",
      "model": "gpt-5.6-luna",
      "variantLabel": "mini-swe-agent - gpt-5-6-luna (high)",
      "score": 44.24778761061947,
      "secondary": {
        "agentHarness": "mini-swe-agent",
        "reasoningEffort": "high",
        "ciMinus": 2.9195672479165324,
        "ciPlus": 2.9195672479165324,
        "ciMethod": "95% run-to-run: SE across repeated whole-benchmark passes (1.96 * std(runs)/sqrt(R))",
        "meanCostUsdPerTask": 0.7779007499999999,
        "medianCostPerTaskUsd": 0.6635542999999999,
        "attempted": 452,
        "passed": 200,
        "tasksAttempted": 113,
        "runs": 4,
        "config": "mini_swe_agent_gpt_5_6_luna_high",
        "costBasis": null
      },
      "source": "https://deepswe.datacurve.ai/artifacts/v1.1/leaderboard-live.json",
      "checkedAt": "2026-09-12"
    },
    {
      "boardId": "deepswe_1_1",
      "model": "glm-5.2",
      "variantLabel": "mini-swe-agent - glm-5-2 (max)",
      "score": 43.77777777777778,
      "secondary": {
        "agentHarness": "mini-swe-agent",
        "reasoningEffort": "max",
        "ciMinus": 1.7256296205808113,
        "ciPlus": 1.7256296205808113,
        "ciMethod": "95% run-to-run: SE across repeated whole-benchmark passes (1.96 * std(runs)/sqrt(R))",
        "meanCostUsdPerTask": 3.9199075743111114,
        "medianCostPerTaskUsd": 3.4637479200000003,
        "attempted": 450,
        "passed": 197,
        "tasksAttempted": 113,
        "runs": 4,
        "config": "mini_swe_agent_glm_5_2_max",
        "costBasis": null
      },
      "source": "https://deepswe.datacurve.ai/artifacts/v1.1/leaderboard-live.json",
      "checkedAt": "2026-09-12"
    },
    {
      "boardId": "deepswe_1_1",
      "model": "grok-4.6",
      "variantLabel": "mini-swe-agent - grok-4-6 (low)",
      "score": 41.648106904231625,
      "secondary": {
        "agentHarness": "mini-swe-agent",
        "reasoningEffort": "low",
        "ciMinus": 2.3186517644299016,
        "ciPlus": 2.3186517644299016,
        "ciMethod": "95% run-to-run: SE across repeated whole-benchmark passes (1.96 * std(runs)/sqrt(R))",
        "meanCostUsdPerTask": 1.0423604231625836,
        "medianCostPerTaskUsd": 0.742614,
        "attempted": 449,
        "passed": 187,
        "tasksAttempted": 113,
        "runs": 4,
        "config": "mini_swe_agent_grok_4_6_low",
        "costBasis": null
      },
      "source": "https://deepswe.datacurve.ai/artifacts/v1.1/leaderboard-live.json",
      "checkedAt": "2026-09-12"
    },
    {
      "boardId": "deepswe_1_1",
      "model": "claude-opus-4.8",
      "variantLabel": "mini-swe-agent - claude-opus-4-8 (low)",
      "score": 40.79822616407982,
      "secondary": {
        "agentHarness": "mini-swe-agent",
        "reasoningEffort": "low",
        "ciMinus": 1.463584758569155,
        "ciPlus": 1.463584758569155,
        "ciMethod": "95% run-to-run: SE across repeated whole-benchmark passes (1.96 * std(runs)/sqrt(R))",
        "meanCostUsdPerTask": 2.293369216740577,
        "medianCostPerTaskUsd": 1.8240152500000004,
        "attempted": 451,
        "passed": 184,
        "tasksAttempted": 113,
        "runs": 4,
        "config": "mini_swe_agent_claude_opus_4_8_low",
        "costBasis": null
      },
      "source": "https://deepswe.datacurve.ai/artifacts/v1.1/leaderboard-live.json",
      "checkedAt": "2026-09-12"
    },
    {
      "boardId": "deepswe_1_1",
      "model": "claude-sonnet-5",
      "variantLabel": "mini-swe-agent - claude-sonnet-5 (medium)",
      "score": 39.77777777777778,
      "secondary": {
        "agentHarness": "mini-swe-agent",
        "reasoningEffort": "medium",
        "ciMinus": 3.127576206329728,
        "ciPlus": 3.127576206329728,
        "ciMethod": "95% run-to-run: SE across repeated whole-benchmark passes (1.96 * std(runs)/sqrt(R))",
        "meanCostUsdPerTask": 4.079036229,
        "medianCostPerTaskUsd": 3.3573110999999995,
        "attempted": 450,
        "passed": 179,
        "tasksAttempted": 113,
        "runs": 4,
        "config": "mini_swe_agent_claude_sonnet_5_medium",
        "costBasis": null
      },
      "source": "https://deepswe.datacurve.ai/artifacts/v1.1/leaderboard-live.json",
      "checkedAt": "2026-09-12"
    },
    {
      "boardId": "deepswe_1_1",
      "model": "glm-5.2",
      "variantLabel": "mini-swe-agent - glm-5-2 (high)",
      "score": 36.283185840707965,
      "secondary": {
        "agentHarness": "mini-swe-agent",
        "reasoningEffort": "high",
        "ciMinus": 4.750072947921659,
        "ciPlus": 4.750072947921659,
        "ciMethod": "95% run-to-run: SE across repeated whole-benchmark passes (1.96 * std(runs)/sqrt(R))",
        "meanCostUsdPerTask": 2.8355164725663715,
        "medianCostPerTaskUsd": 2.25012998,
        "attempted": 452,
        "passed": 164,
        "tasksAttempted": 113,
        "runs": 4,
        "config": "mini_swe_agent_glm_5_2_high",
        "costBasis": null
      },
      "source": "https://deepswe.datacurve.ai/artifacts/v1.1/leaderboard-live.json",
      "checkedAt": "2026-09-12"
    },
    {
      "boardId": "deepswe_1_1",
      "model": "gemini-3.5-flash",
      "variantLabel": "mini-swe-agent - gemini-3-5-flash (high)",
      "score": 36.06194690265487,
      "secondary": {
        "agentHarness": "mini-swe-agent",
        "reasoningEffort": "high",
        "ciMinus": 3.9663030105204413,
        "ciPlus": 3.9663030105204413,
        "ciMethod": "95% run-to-run: SE across repeated whole-benchmark passes (1.96 * std(runs)/sqrt(R))",
        "meanCostUsdPerTask": 3.4467114588691796,
        "medianCostPerTaskUsd": 2.8624087500000015,
        "attempted": 452,
        "passed": 163,
        "tasksAttempted": 113,
        "runs": 4,
        "config": "mini_swe_agent_gemini_3_5_flash_high",
        "costBasis": null
      },
      "source": "https://deepswe.datacurve.ai/artifacts/v1.1/leaderboard-live.json",
      "checkedAt": "2026-09-12"
    },
    {
      "boardId": "deepswe_1_1",
      "model": "gpt-5.6-terra",
      "variantLabel": "mini-swe-agent - gpt-5-6-terra (medium)",
      "score": 35.11111111111111,
      "secondary": {
        "agentHarness": "mini-swe-agent",
        "reasoningEffort": "medium",
        "ciMinus": 3.3819643034066296,
        "ciPlus": 3.381964303406633,
        "ciMethod": "95% run-to-run: SE across repeated whole-benchmark passes (1.96 * std(runs)/sqrt(R))",
        "meanCostUsdPerTask": 0.5832493688888889,
        "medianCostPerTaskUsd": 0.546099,
        "attempted": 450,
        "passed": 158,
        "tasksAttempted": 113,
        "runs": 4,
        "config": "mini_swe_agent_gpt_5_6_terra_medium",
        "costBasis": null
      },
      "source": "https://deepswe.datacurve.ai/artifacts/v1.1/leaderboard-live.json",
      "checkedAt": "2026-09-12"
    },
    {
      "boardId": "deepswe_1_1",
      "model": "kimi-k2.7-code",
      "variantLabel": "mini-swe-agent - kimi-k2-7-code (None)",
      "score": 30.53097345132743,
      "secondary": {
        "agentHarness": "mini-swe-agent",
        "reasoningEffort": null,
        "ciMinus": 0.500701652246093,
        "ciPlus": 0.5007016522460965,
        "ciMethod": "95% run-to-run: SE across repeated whole-benchmark passes (1.96 * std(runs)/sqrt(R))",
        "meanCostUsdPerTask": 2.815536243274336,
        "medianCostPerTaskUsd": 2.1955079100000003,
        "attempted": 452,
        "passed": 138,
        "tasksAttempted": 113,
        "runs": 4,
        "config": "mini_swe_agent_kimi_k2_7_code_default",
        "costBasis": null
      },
      "source": "https://deepswe.datacurve.ai/artifacts/v1.1/leaderboard-live.json",
      "checkedAt": "2026-09-12"
    },
    {
      "boardId": "deepswe_1_1",
      "model": "claude-sonnet-5",
      "variantLabel": "mini-swe-agent - claude-sonnet-5 (low)",
      "score": 30.51224944320713,
      "secondary": {
        "agentHarness": "mini-swe-agent",
        "reasoningEffort": "low",
        "ciMinus": 1.1260175414427458,
        "ciPlus": 1.1260175414427493,
        "ciMethod": "95% run-to-run: SE across repeated whole-benchmark passes (1.96 * std(runs)/sqrt(R))",
        "meanCostUsdPerTask": 2.1865550832962137,
        "medianCostPerTaskUsd": 1.6764955500000003,
        "attempted": 449,
        "passed": 137,
        "tasksAttempted": 113,
        "runs": 4,
        "config": "mini_swe_agent_claude_sonnet_5_low",
        "costBasis": null
      },
      "source": "https://deepswe.datacurve.ai/artifacts/v1.1/leaderboard-live.json",
      "checkedAt": "2026-09-12"
    },
    {
      "boardId": "deepswe_1_1",
      "model": "claude-sonnet-4.6",
      "variantLabel": "mini-swe-agent - claude-sonnet-4-6 (high)",
      "score": 29.933481152993345,
      "secondary": {
        "agentHarness": "mini-swe-agent",
        "reasoningEffort": "high",
        "ciMinus": 4.092141269189135,
        "ciPlus": 4.092141269189142,
        "ciMethod": "95% run-to-run: SE across repeated whole-benchmark passes (1.96 * std(runs)/sqrt(R))",
        "meanCostUsdPerTask": 5.5223517987804875,
        "medianCostPerTaskUsd": 4.867440150000001,
        "attempted": 451,
        "passed": 135,
        "tasksAttempted": 113,
        "runs": 4,
        "config": "mini_swe_agent_claude_sonnet_4_6_high",
        "costBasis": null
      },
      "source": "https://deepswe.datacurve.ai/artifacts/v1.1/leaderboard-live.json",
      "checkedAt": "2026-09-12"
    },
    {
      "boardId": "deepswe_1_1",
      "model": "gpt-5.5",
      "variantLabel": "mini-swe-agent - gpt-5-5 (low)",
      "score": 26.991150442477874,
      "secondary": {
        "agentHarness": "mini-swe-agent",
        "reasoningEffort": "low",
        "ciMinus": 2.2945032220071795,
        "ciPlus": 2.294503222007183,
        "ciMethod": "95% run-to-run: SE across repeated whole-benchmark passes (1.96 * std(runs)/sqrt(R))",
        "meanCostUsdPerTask": 1.2002025442477877,
        "medianCostPerTaskUsd": 1.108106,
        "attempted": 452,
        "passed": 122,
        "tasksAttempted": 113,
        "runs": 4,
        "config": "mini_swe_agent_gpt_5_5_low",
        "costBasis": null
      },
      "source": "https://deepswe.datacurve.ai/artifacts/v1.1/leaderboard-live.json",
      "checkedAt": "2026-09-12"
    },
    {
      "boardId": "deepswe_1_1",
      "model": "gpt-5.6-terra",
      "variantLabel": "mini-swe-agent - gpt-5-6-terra (low)",
      "score": 24.053452115812917,
      "secondary": {
        "agentHarness": "mini-swe-agent",
        "reasoningEffort": "low",
        "ciMinus": 0.7767613669447826,
        "ciPlus": 0.7767613669447861,
        "ciMethod": "95% run-to-run: SE across repeated whole-benchmark passes (1.96 * std(runs)/sqrt(R))",
        "meanCostUsdPerTask": 0.4277452605790646,
        "medianCostPerTaskUsd": 0.3989895,
        "attempted": 449,
        "passed": 108,
        "tasksAttempted": 113,
        "runs": 4,
        "config": "mini_swe_agent_gpt_5_6_terra_low",
        "costBasis": null
      },
      "source": "https://deepswe.datacurve.ai/artifacts/v1.1/leaderboard-live.json",
      "checkedAt": "2026-09-12"
    },
    {
      "boardId": "deepswe_1_1",
      "model": "gemini-3.1-pro-preview",
      "variantLabel": "mini-swe-agent - gemini-3-1-pro-preview (high)",
      "score": 11.725663716814159,
      "secondary": {
        "agentHarness": "mini-swe-agent",
        "reasoningEffort": "high",
        "ciMinus": 1.4810954611088434,
        "ciPlus": 1.481095461108847,
        "ciMethod": "95% run-to-run: SE across repeated whole-benchmark passes (1.96 * std(runs)/sqrt(R))",
        "meanCostUsdPerTask": 2.1434045568888886,
        "medianCostPerTaskUsd": 1.7235064000000002,
        "attempted": 452,
        "passed": 53,
        "tasksAttempted": 113,
        "runs": 4,
        "config": "mini_swe_agent_gemini_3_1_pro_preview_high",
        "costBasis": null
      },
      "source": "https://deepswe.datacurve.ai/artifacts/v1.1/leaderboard-live.json",
      "checkedAt": "2026-09-12"
    },
    {
      "boardId": "deepswe_1_1",
      "model": "gpt-5.6-luna",
      "variantLabel": "mini-swe-agent - gpt-5-6-luna (medium)",
      "score": 11.283185840707963,
      "secondary": {
        "agentHarness": "mini-swe-agent",
        "reasoningEffort": "medium",
        "ciMinus": 0.8303197562056504,
        "ciPlus": 0.8303197562056521,
        "ciMethod": "95% run-to-run: SE across repeated whole-benchmark passes (1.96 * std(runs)/sqrt(R))",
        "meanCostUsdPerTask": 0.21630978982300886,
        "medianCostPerTaskUsd": 0.1994702,
        "attempted": 452,
        "passed": 51,
        "tasksAttempted": 113,
        "runs": 4,
        "config": "mini_swe_agent_gpt_5_6_luna_medium",
        "costBasis": null
      },
      "source": "https://deepswe.datacurve.ai/artifacts/v1.1/leaderboard-live.json",
      "checkedAt": "2026-09-12"
    },
    {
      "boardId": "deepswe_1_1",
      "model": "gpt-5.6-luna",
      "variantLabel": "mini-swe-agent - gpt-5-6-luna (low)",
      "score": 1.5486725663716814,
      "secondary": {
        "agentHarness": "mini-swe-agent",
        "reasoningEffort": "low",
        "ciMinus": 0.8303197562056515,
        "ciPlus": 0.8303197562056512,
        "ciMethod": "95% run-to-run: SE across repeated whole-benchmark passes (1.96 * std(runs)/sqrt(R))",
        "meanCostUsdPerTask": 0.07240623938053097,
        "medianCostPerTaskUsd": 0.0697851,
        "attempted": 452,
        "passed": 7,
        "tasksAttempted": 113,
        "runs": 4,
        "config": "mini_swe_agent_gpt_5_6_luna_low",
        "costBasis": null
      },
      "source": "https://deepswe.datacurve.ai/artifacts/v1.1/leaderboard-live.json",
      "checkedAt": "2026-09-12"
    }
  ]
}