{
  "api_version": "1",
  "updated": "2026-10-29",
  "latest_event": "2026-10-28",
  "rights": "© Frontier Log, all rights reserved.",
  "benchmark": {
    "slug": "agents-last-exam",
    "name": "Agents' Last Exam v1",
    "family": "agents-last-exam",
    "unit": "%",
    "higher_is_better": true,
    "status": "superseded",
    "active": false,
    "superseded_by": "agents-last-exam-1-1",
    "retired": null,
    "benchmark_url": "https://agents-last-exam.org",
    "introduced": "2026-06-03",
    "creator": "UC Berkeley RDI",
    "introduced_source": "https://arxiv.org/abs/2606.05405v1",
    "settings": [
      {
        "setting": "pass",
        "name": "Pass rate",
        "url": "https://frontierlog.ai/benchmark/agents-last-exam",
        "results": 8
      },
      {
        "setting": "score",
        "name": "Score",
        "url": "https://frontierlog.ai/benchmark/agents-last-exam/score",
        "results": 4
      },
      {
        "setting": "cli-pass",
        "name": "CLI subset, pass rate",
        "url": "https://frontierlog.ai/benchmark/agents-last-exam/cli-pass",
        "results": 8
      },
      {
        "setting": "unstated",
        "name": "Measure not stated",
        "url": "https://frontierlog.ai/benchmark/agents-last-exam/unstated",
        "results": 8
      }
    ]
  },
  "total": 8,
  "results": [
    {
      "id": "gpt-5-5:agents-last-exam:unstated:1",
      "model": "gpt-5-5",
      "model_name": "GPT-5.5",
      "model_date": "2026-04-23",
      "event": "https://frontierlog.ai/event/2026-04-23-gpt-5-5",
      "benchmark": "agents-last-exam",
      "benchmark_name": "Agents' Last Exam v1",
      "setting": "unstated",
      "setting_name": "Measure not stated",
      "score": 46.9,
      "unit": "%",
      "printed": "46.9%",
      "reported_by": "openai",
      "run_by": null,
      "label": "GPT‑5.5",
      "source": "https://openai.com/index/gpt-5-6",
      "source_title": "GPT-5.6: Frontier intelligence that scales with your ambition",
      "source_date": "2026-07-09",
      "where": "later report: GPT-5.6 post, professional eval table (first table), row 'Agents' Last Exam', column 'GPT-5.5'",
      "own": false,
      "primary": true,
      "parallel": false,
      "notes": "effort and harness not stated",
      "caveat": "",
      "human_verified": false,
      "human_verified_on": null,
      "check": "double_entry",
      "corroborated": false,
      "corroborated_on": null,
      "corroborated_by": []
    },
    {
      "id": "glm-5-2:agents-last-exam:unstated:1",
      "model": "glm-5-2",
      "model_name": "GLM-5.2",
      "model_date": "2026-06-16",
      "event": "https://frontierlog.ai/event/2026-06-16-glm-5-2",
      "benchmark": "agents-last-exam",
      "benchmark_name": "Agents' Last Exam v1",
      "setting": "unstated",
      "setting_name": "Measure not stated",
      "score": 20.4,
      "unit": "%",
      "printed": "20.4",
      "reported_by": "zai",
      "run_by": null,
      "label": "GLM-5.2",
      "source": "https://z.ai/blog/glm-5.3-flash",
      "source_title": "GLM-5.3-Flash: Frontier Intelligence, Flash Cost",
      "source_date": "2026-08-26",
      "where": "later report: GLM-5.3-Flash post, main benchmark table (second table), AGENTIC block, row 'Agents' Last Exam', column 'GLM-5.2'",
      "own": false,
      "primary": true,
      "parallel": false,
      "notes": "Claude Code harness; subset not stated",
      "caveat": "",
      "human_verified": false,
      "human_verified_on": null,
      "check": "corroborated",
      "corroborated": true,
      "corroborated_on": "2026-10-29",
      "corroborated_by": [
        {
          "url": "https://huggingface.co/zai-org/GLM-5.3-Flash",
          "name": "GLM-5.3-Flash card",
          "title": "zai-org/GLM-5.3-Flash model card",
          "lab": "zai",
          "date": "2026-08-26",
          "printed": "20.4",
          "where": "bench_53.png chart 'LLM Performance Evaluation' (subtitle 'Agents' Last Exam', no subset), panel 'Agents' Last Exam', GLM-5.2 bar"
        },
        {
          "url": "https://github.com/zai-org/GLM-5",
          "name": "GLM-5 card",
          "title": "GLM-5: From Vibe Coding to Agentic Engineering (zai-org/GLM-5 GitHub repository)",
          "lab": "zai",
          "date": "2026-02-12",
          "printed": "20.4",
          "where": "README, image resources/bench_53.png (GLM-5.3-Flash chart), Agents' Last Exam panel, GLM-5.2 bar"
        },
        {
          "url": "https://docs.z.ai/guides/llm/glm-5.3-flash",
          "name": "GLM-5.3-Flash card",
          "title": "GLM-5.3-Flash/FlashX - Overview",
          "lab": "zai",
          "date": "2026-08-26",
          "printed": "20.4",
          "where": "Benchmark image (GLM-5.3-Flash chart), panel 'Agents' Last Exam', GLM-5.2 bar"
        }
      ],
      "source_capture": "https://frontierlog.ai/captures/58a995e241d3e40c.png"
    },
    {
      "id": "gpt-5-6-luna:agents-last-exam:unstated:1",
      "model": "gpt-5-6-luna",
      "model_name": "GPT-5.6 Luna",
      "model_date": "2026-07-09",
      "event": "https://frontierlog.ai/event/2026-07-09-gpt-5-6",
      "benchmark": "agents-last-exam",
      "benchmark_name": "Agents' Last Exam v1",
      "setting": "unstated",
      "setting_name": "Measure not stated",
      "score": 50.3,
      "unit": "%",
      "printed": "50.3%",
      "reported_by": "openai",
      "run_by": null,
      "label": "GPT‑5.6 Luna",
      "source": "https://openai.com/index/gpt-5-6",
      "source_title": "GPT-5.6: Frontier intelligence that scales with your ambition",
      "source_date": "2026-07-09",
      "where": "professional eval table (first table), row 'Agents' Last Exam', column 'GPT-5.6 Luna'",
      "own": true,
      "primary": true,
      "parallel": false,
      "notes": "effort and harness not stated",
      "caveat": "",
      "human_verified": false,
      "human_verified_on": null,
      "check": "double_entry",
      "corroborated": false,
      "corroborated_on": null,
      "corroborated_by": []
    },
    {
      "id": "gpt-5-6-sol:agents-last-exam:unstated:1",
      "model": "gpt-5-6-sol",
      "model_name": "GPT-5.6 Sol",
      "model_date": "2026-07-09",
      "event": "https://frontierlog.ai/event/2026-07-09-gpt-5-6",
      "benchmark": "agents-last-exam",
      "benchmark_name": "Agents' Last Exam v1",
      "setting": "unstated",
      "setting_name": "Measure not stated",
      "score": 53.6,
      "unit": "%",
      "printed": "53.6",
      "reported_by": "openai",
      "run_by": null,
      "label": "GPT-5.6 Sol",
      "source": "https://openai.com/index/gpt-5-6",
      "source_title": "GPT-5.6: Frontier intelligence that scales with your ambition",
      "source_date": "2026-07-09",
      "where": "opening section, paragraph beginning 'We trained GPT-5.6 to get more useful work from every token', the GPT-5.6 Sol figure on Agents' Last Exam",
      "own": true,
      "primary": true,
      "parallel": false,
      "notes": "headline figure; effort and harness not stated",
      "caveat": "",
      "human_verified": false,
      "human_verified_on": null,
      "check": "corroborated",
      "corroborated": true,
      "corroborated_on": "2026-10-29",
      "corroborated_by": [
        {
          "url": "https://qwen.ai/blog?id=qwen3.8",
          "name": "Qwen3.8-Max",
          "title": "Qwen3.8-Max: A New Bar for Coding and Cowork",
          "lab": "alibaba",
          "date": "2026-08-03",
          "printed": "53.6",
          "where": "Alibaba (another lab's table): performance chart, panel \"Agents' Last Exam (Score)\", OpenAI-logo bar (legend \"GPT5.6 Sol (max)\")"
        },
        {
          "url": "https://openai.com/index/gpt-6-astra",
          "name": "GPT-6 Astra",
          "title": "GPT-6 Astra: A new generation of intelligence",
          "lab": "openai",
          "date": "2026-09-03",
          "printed": "53.6%",
          "where": "Computer Use eval table (HTML), row 'Agents' Last Exam', column 'GPT‑5.6 Sol' (header carries footnote 2); also the sentence '...compared with 55.5% for Claude Opus 5 and 53.6% for GPT‑5.6 Sol' (highest-scoring settings)"
        }
      ]
    },
    {
      "id": "gpt-5-6-sol:agents-last-exam:unstated:2",
      "model": "gpt-5-6-sol",
      "model_name": "GPT-5.6 Sol",
      "model_date": "2026-07-09",
      "event": "https://frontierlog.ai/event/2026-07-09-gpt-5-6",
      "benchmark": "agents-last-exam",
      "benchmark_name": "Agents' Last Exam v1",
      "setting": "unstated",
      "setting_name": "Measure not stated",
      "score": 52.7,
      "unit": "%",
      "printed": "52.7%",
      "reported_by": "openai",
      "run_by": null,
      "label": "GPT‑5.6 Sol",
      "source": "https://openai.com/index/gpt-5-6",
      "source_title": "GPT-5.6: Frontier intelligence that scales with your ambition",
      "source_date": "2026-07-09",
      "where": "professional eval table (first table, columns GPT-5.6 Sol, Terra, Luna, GPT-5.5, ...), row 'Agents' Last Exam', column 'GPT-5.6 Sol'",
      "own": true,
      "primary": false,
      "parallel": false,
      "notes": "eval table figure; effort and harness not stated",
      "caveat": "",
      "human_verified": false,
      "human_verified_on": null,
      "check": "double_entry",
      "corroborated": false,
      "corroborated_on": null,
      "corroborated_by": []
    },
    {
      "id": "gpt-5-6-terra:agents-last-exam:unstated:1",
      "model": "gpt-5-6-terra",
      "model_name": "GPT-5.6 Terra",
      "model_date": "2026-07-09",
      "event": "https://frontierlog.ai/event/2026-07-09-gpt-5-6",
      "benchmark": "agents-last-exam",
      "benchmark_name": "Agents' Last Exam v1",
      "setting": "unstated",
      "setting_name": "Measure not stated",
      "score": 50.4,
      "unit": "%",
      "printed": "50.4%",
      "reported_by": "openai",
      "run_by": null,
      "label": "GPT‑5.6 Terra",
      "source": "https://openai.com/index/gpt-5-6",
      "source_title": "GPT-5.6: Frontier intelligence that scales with your ambition",
      "source_date": "2026-07-09",
      "where": "professional eval table (first table), row 'Agents' Last Exam', column 'GPT-5.6 Terra'",
      "own": true,
      "primary": true,
      "parallel": false,
      "notes": "effort and harness not stated",
      "caveat": "",
      "human_verified": false,
      "human_verified_on": null,
      "check": "double_entry",
      "corroborated": false,
      "corroborated_on": null,
      "corroborated_by": []
    },
    {
      "id": "glm-5-3-flash:agents-last-exam:unstated:1",
      "model": "glm-5-3-flash",
      "model_name": "GLM-5.3-Flash",
      "model_date": "2026-08-26",
      "event": "https://frontierlog.ai/event/2026-08-26-glm-5-3-flash",
      "benchmark": "agents-last-exam",
      "benchmark_name": "Agents' Last Exam v1",
      "setting": "unstated",
      "setting_name": "Measure not stated",
      "score": 26.3,
      "unit": "%",
      "printed": "26.3",
      "reported_by": "zai",
      "run_by": null,
      "label": "GLM-5.3-Flash",
      "source": "https://z.ai/blog/glm-5.3-flash",
      "source_title": "GLM-5.3-Flash: Frontier Intelligence, Flash Cost",
      "source_date": "2026-08-26",
      "where": "main benchmark table (second table), AGENTIC block, row 'Agents' Last Exam', column 'GLM-5.3-Flash'",
      "own": true,
      "primary": true,
      "parallel": false,
      "notes": "Claude Code harness; max effort; 1M context; subset not stated",
      "caveat": "",
      "human_verified": false,
      "human_verified_on": null,
      "check": "corroborated",
      "corroborated": true,
      "corroborated_on": "2026-10-29",
      "corroborated_by": [
        {
          "url": "https://huggingface.co/zai-org/GLM-5.3-Flash",
          "name": "GLM-5.3-Flash card",
          "title": "zai-org/GLM-5.3-Flash model card",
          "lab": "zai",
          "date": "2026-08-26",
          "printed": "26.3",
          "where": "bench_53.png chart 'LLM Performance Evaluation' (subtitle 'Agents' Last Exam', no subset), panel 'Agents' Last Exam', GLM-5.3-Flash bar"
        },
        {
          "url": "https://github.com/zai-org/GLM-5",
          "name": "GLM-5 card",
          "title": "GLM-5: From Vibe Coding to Agentic Engineering (zai-org/GLM-5 GitHub repository)",
          "lab": "zai",
          "date": "2026-02-12",
          "printed": "26.3",
          "where": "README, image resources/bench_53.png, Agents' Last Exam panel, GLM-5.3-Flash bar"
        },
        {
          "url": "https://docs.z.ai/guides/llm/glm-5.3-flash",
          "name": "GLM-5.3-Flash card",
          "title": "GLM-5.3-Flash/FlashX - Overview",
          "lab": "zai",
          "date": "2026-08-26",
          "printed": "26.3",
          "where": "Benchmark image, panel 'Agents' Last Exam', GLM-5.3-Flash bar"
        }
      ],
      "source_capture": "https://frontierlog.ai/captures/58a995e241d3e40c.png"
    },
    {
      "id": "gpt-6-astra:agents-last-exam:unstated:1",
      "model": "gpt-6-astra",
      "model_name": "GPT-6 Astra",
      "model_date": "2026-09-03",
      "event": "https://frontierlog.ai/event/2026-09-03-gpt-6-astra",
      "benchmark": "agents-last-exam",
      "benchmark_name": "Agents' Last Exam v1",
      "setting": "unstated",
      "setting_name": "Measure not stated",
      "score": 59.3,
      "unit": "%",
      "printed": "59.3%",
      "reported_by": "openai",
      "run_by": null,
      "label": "GPT‑6 Astra",
      "source": "https://openai.com/index/gpt-6-astra",
      "source_title": "GPT-6 Astra: A new generation of intelligence",
      "source_date": "2026-09-03",
      "where": "'Computer Use' eval table, row 'Agents' Last Exam', column 'GPT-6 Astra'",
      "own": true,
      "primary": true,
      "parallel": false,
      "notes": "best of any effort; harness not stated",
      "caveat": "",
      "human_verified": false,
      "human_verified_on": null,
      "check": "double_entry",
      "corroborated": false,
      "corroborated_on": null,
      "corroborated_by": []
    }
  ]
}