{
  "api_version": "1",
  "updated": "2026-11-05",
  "latest_event": "2026-11-04",
  "rights": "© Frontier Log, all rights reserved.",
  "event": {
    "slug": "2026-11-04-openai-reports-astra-underperformance",
    "date": "2026-11-04",
    "provider": "openai",
    "importance": "minor",
    "title": "OpenAI reports Astra's covert underperformance",
    "headline": "OpenAI reports Astra's covert underperformance",
    "detail": "On safety-framed tasks Astra completed about 10% of rubric items, against about 16% framed as capabilities.",
    "tags": [
      "policy",
      "safety"
    ],
    "related": [],
    "source": "https://alignment.openai.com/misalignment-reports/covert-underperformance-on-safety-research-tasks/",
    "models": [
      {
        "slug": "gpt-6-astra",
        "name": "GPT-6 Astra",
        "lab": "openai",
        "date": "2026-09-03",
        "source": "https://openai.com/index/gpt-6-astra",
        "source_title": "GPT-6 Astra: A new generation of intelligence",
        "flagship": true,
        "event": "https://frontierlog.ai/event/2026-09-03-gpt-6-astra",
        "releases": [],
        "variant_of": null,
        "made_by": "OpenAI",
        "url": "https://frontierlog.ai/model/gpt-6-astra",
        "results": [
          {
            "id": "gpt-6-astra:terminal-bench-4-0:default: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": "terminal-bench-4-0",
            "benchmark_name": "Terminal-Bench 4.0",
            "setting": "default",
            "setting_name": "",
            "score": 57.9,
            "unit": "%",
            "printed": "57.9%",
            "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": "Coding table, row 'Terminal-Bench 4.0', column 'GPT‑6 Astra'",
            "own": true,
            "primary": true,
            "parallel": false,
            "notes": "",
            "caveat": "",
            "human_verified": true,
            "human_verified_on": "2026-10-16",
            "check": "human",
            "corroborated": true,
            "corroborated_on": "2026-10-19",
            "corroborated_by": [
              {
                "url": "https://www.anthropic.com/claude-haiku-5-5-system-card",
                "name": "Claude Haiku 5.5 card",
                "title": "Claude Haiku 5.5 System Card",
                "lab": "anthropic",
                "date": "2026-10-07",
                "printed": "57.9%",
                "where": "Section 8.4 (p.115-116) sentence: \"OpenAI reported GPT-6 Astra's final Terminal-Bench 4.0 score at 57.9%, scored at high thinking effort\""
              },
              {
                "url": "https://www.anthropic.com/claude-opus-5-5-system-card",
                "name": "Claude Opus 5.5 card",
                "title": "Claude Opus 5.5 System Card",
                "lab": "anthropic",
                "date": "2026-09-22",
                "printed": "57.9",
                "where": "Table 8.1.A (p.174), row Terminal-Bench 4.0, column GPT-6 Astra"
              },
              {
                "url": "https://www.anthropic.com/claude-sonnet-5-5-system-card",
                "name": "Claude Sonnet 5.5 card",
                "title": "Claude Sonnet 5.5 System Card",
                "lab": "anthropic",
                "date": "2026-09-28",
                "printed": "57.9%",
                "where": "Section 8.5 (p.113) sentence: \"GPT-6 Astra reported their final Terminal-Bench 4.0 score at 57.9%, scored at high thinking effort\""
              },
              {
                "url": "https://www.anthropic.com/claude-opus-5-5",
                "name": "Claude Opus 5.5",
                "title": "Introducing Claude Opus 5.5",
                "lab": "anthropic",
                "date": "2026-09-22",
                "printed": "57.9%",
                "where": "HTML benchmark table under 'Performance and cost-effectiveness', row Terminal-Bench 4.0¹, column GPT-6 Astra (high effort, as reported by OpenAI)"
              }
            ],
            "source_capture": "https://frontierlog.ai/captures/4de8f321fbadf8b8.png"
          },
          {
            "id": "gpt-6-astra:terminal-bench-science-0-1:default: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": "terminal-bench-science-0-1",
            "benchmark_name": "Terminal-Bench Science 0.1",
            "setting": "default",
            "setting_name": "",
            "score": 64.6,
            "unit": "%",
            "printed": "64.6%",
            "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": "Academic table, row 'Terminal-Bench Science 0.1', column 'GPT‑6 Astra'",
            "own": true,
            "primary": true,
            "parallel": false,
            "notes": "highest score at any effort",
            "caveat": "",
            "human_verified": true,
            "human_verified_on": "2026-10-16",
            "check": "human",
            "corroborated": true,
            "corroborated_on": "2026-10-19",
            "corroborated_by": [
              {
                "url": "https://www.anthropic.com/claude-opus-5-5-system-card",
                "name": "Claude Opus 5.5 card",
                "title": "Claude Opus 5.5 System Card",
                "lab": "anthropic",
                "date": "2026-09-22",
                "printed": "64.6",
                "where": "Table 8.1.A (p.174), row Terminal-Bench-Science 0.1, column GPT-6 Astra"
              },
              {
                "url": "https://www.anthropic.com/claude-sonnet-5-5-system-card",
                "name": "Claude Sonnet 5.5 card",
                "title": "Claude Sonnet 5.5 System Card",
                "lab": "anthropic",
                "date": "2026-09-28",
                "printed": "64.6%",
                "where": "Section 8.6 (p.114) sentence: \"OpenAI reported GPT-6 Astra at 64.6% with max thinking effort\""
              },
              {
                "url": "https://www.anthropic.com/claude-opus-5-5",
                "name": "Claude Opus 5.5",
                "title": "Introducing Claude Opus 5.5",
                "lab": "anthropic",
                "date": "2026-09-22",
                "printed": "64.6%",
                "where": "HTML benchmark table under 'Performance and cost-effectiveness', row Terminal-Bench-Science 0.1³, column GPT-6 Astra (as reported by OpenAI)"
              }
            ],
            "source_capture": "https://frontierlog.ai/captures/478a8e3e7a393b2d.png"
          },
          {
            "id": "gpt-6-astra:automationbench:default: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": "automationbench",
            "benchmark_name": "AutomationBench",
            "setting": "default",
            "setting_name": "",
            "score": 41.4,
            "unit": "%",
            "printed": "41.4%",
            "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": "Professional table, row 'AutomationBench', column 'GPT‑6 Astra'",
            "own": true,
            "primary": true,
            "parallel": false,
            "notes": "highest score at any effort",
            "caveat": "",
            "human_verified": true,
            "human_verified_on": "2026-10-16",
            "check": "human",
            "corroborated": true,
            "corroborated_on": "2026-10-19",
            "corroborated_by": [
              {
                "url": "https://www.anthropic.com/claude-opus-5-5-system-card",
                "name": "Claude Opus 5.5 card",
                "title": "Claude Opus 5.5 System Card",
                "lab": "anthropic",
                "date": "2026-09-22",
                "printed": "41.4",
                "where": "Table 8.1.A (p.174), row AutomationBench, column GPT-6 Astra"
              },
              {
                "url": "https://www.anthropic.com/claude-opus-5-5",
                "name": "Claude Opus 5.5",
                "title": "Introducing Claude Opus 5.5",
                "lab": "anthropic",
                "date": "2026-09-22",
                "printed": "41.4%",
                "where": "HTML benchmark table under 'Performance and cost-effectiveness', row AutomationBench² (run and reported by Zapier), column GPT-6 Astra"
              },
              {
                "url": "https://blog.google/innovation-and-ai/models-and-research/gemini-models/gemini-4-argon/",
                "name": "Gemini 4 Argon",
                "title": "Gemini 4 Argon: our next era of frontier intelligence",
                "lab": "google",
                "date": "2026-09-30",
                "printed": "41.4%",
                "where": "Benchmark table image; row 'AutomationBench, Score'; column gpt-6-astra"
              }
            ],
            "source_capture": "https://frontierlog.ai/captures/8618c81cd990eeed.png"
          },
          {
            "id": "gpt-6-astra:osworld-2-0:offline-partial: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": "osworld-2-0",
            "benchmark_name": "OSWorld 2.0",
            "setting": "offline-partial",
            "setting_name": "Offline set, partial score",
            "score": 72.6,
            "unit": "%",
            "printed": "72.6%",
            "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 table, row 'OSWorld 2.0 (v2026.08.08, offline set, partial score)', column 'GPT‑6 Astra'",
            "own": true,
            "primary": true,
            "parallel": false,
            "notes": "task release v2026.08.08",
            "caveat": "",
            "human_verified": true,
            "human_verified_on": "2026-10-16",
            "check": "human",
            "corroborated": false,
            "corroborated_on": null,
            "corroborated_by": [],
            "source_capture": "https://frontierlog.ai/captures/fbd1c3587f0ca67d.png"
          },
          {
            "id": "gpt-6-astra:hle:with-tools: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": "hle",
            "benchmark_name": "Humanity's Last Exam",
            "setting": "with-tools",
            "setting_name": "With tools",
            "score": 57.2,
            "unit": "%",
            "printed": "57.2%",
            "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": "Academic table, row 'Humanity's Last Exam (w/ tools)', column 'GPT‑6 Astra'",
            "own": true,
            "primary": true,
            "parallel": false,
            "notes": "",
            "caveat": "",
            "human_verified": true,
            "human_verified_on": "2026-10-16",
            "check": "human",
            "corroborated": true,
            "corroborated_on": "2026-10-19",
            "corroborated_by": [
              {
                "url": "https://www.anthropic.com/claude-opus-5-5-system-card",
                "name": "Claude Opus 5.5 card",
                "title": "Claude Opus 5.5 System Card",
                "lab": "anthropic",
                "date": "2026-09-22",
                "printed": "57.2",
                "where": "Table 8.1.A (p.174), row Humanity's Last Exam / With tools, column GPT-6 Astra"
              },
              {
                "url": "https://www.anthropic.com/claude-opus-5-5",
                "name": "Claude Opus 5.5",
                "title": "Introducing Claude Opus 5.5",
                "lab": "anthropic",
                "date": "2026-09-22",
                "printed": "57.2%",
                "where": "HTML benchmark table under 'Performance and cost-effectiveness', row Humanity's Last Exam (with tools), column GPT-6 Astra"
              }
            ],
            "source_capture": "https://frontierlog.ai/captures/478a8e3e7a393b2d.png"
          },
          {
            "id": "gpt-6-astra:frontiercode-1-1:main: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": "frontiercode-1-1",
            "benchmark_name": "FrontierCode 1.1",
            "setting": "main",
            "setting_name": "Main set",
            "score": 53.3,
            "unit": "%",
            "printed": "53.3%",
            "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": "Coding table, row 'FrontierCode 1.1 Main (score)', column 'GPT‑6 Astra'",
            "own": true,
            "primary": true,
            "parallel": false,
            "notes": "",
            "caveat": "",
            "human_verified": true,
            "human_verified_on": "2026-10-16",
            "check": "human",
            "corroborated": true,
            "corroborated_on": "2026-10-19",
            "corroborated_by": [
              {
                "url": "https://www.anthropic.com/claude-opus-5-5-system-card",
                "name": "Claude Opus 5.5 card",
                "title": "Claude Opus 5.5 System Card",
                "lab": "anthropic",
                "date": "2026-09-22",
                "printed": "53.3",
                "where": "Table 8.1.A (p.174), row FrontierCode v1.1 (Main), column GPT-6 Astra"
              },
              {
                "url": "https://www.anthropic.com/claude-opus-5-5",
                "name": "Claude Opus 5.5",
                "title": "Introducing Claude Opus 5.5",
                "lab": "anthropic",
                "date": "2026-09-22",
                "printed": "53.3%",
                "where": "HTML benchmark table under 'Performance and cost-effectiveness', row FrontierCode v1.1 (Main), column GPT-6 Astra"
              }
            ],
            "source_capture": "https://frontierlog.ai/captures/4de8f321fbadf8b8.png"
          },
          {
            "id": "gpt-6-astra:deepswe-1-1:default: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": "deepswe-1-1",
            "benchmark_name": "DeepSWE v1.1",
            "setting": "default",
            "setting_name": "",
            "score": 74.1,
            "unit": "%",
            "printed": "74.1%",
            "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": "Coding table, row 'DeepSWE v1.1', column 'GPT‑6 Astra'",
            "own": true,
            "primary": true,
            "parallel": false,
            "notes": "",
            "caveat": "",
            "human_verified": true,
            "human_verified_on": "2026-10-16",
            "check": "human",
            "corroborated": true,
            "corroborated_on": "2026-10-19",
            "corroborated_by": [
              {
                "url": "https://blog.google/innovation-and-ai/models-and-research/gemini-models/gemini-4-argon/",
                "name": "Gemini 4 Argon",
                "title": "Gemini 4 Argon: our next era of frontier intelligence",
                "lab": "google",
                "date": "2026-09-30",
                "printed": "74.1%",
                "where": "Benchmark table image; row 'DeepSWE v1.1'; column gpt-6-astra"
              }
            ],
            "source_capture": "https://frontierlog.ai/captures/4de8f321fbadf8b8.png"
          },
          {
            "id": "gpt-6-astra:gpqa-diamond:default: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": "gpqa-diamond",
            "benchmark_name": "GPQA Diamond",
            "setting": "default",
            "setting_name": "",
            "score": 96.0,
            "unit": "%",
            "printed": "96.0%",
            "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": "'Academic' table, column 'GPT‑6 Astra', row 'GPQA Diamond'",
            "own": true,
            "primary": true,
            "parallel": false,
            "notes": "",
            "caveat": "",
            "human_verified": true,
            "human_verified_on": "2026-10-17",
            "check": "human",
            "corroborated": false,
            "corroborated_on": null,
            "corroborated_by": [],
            "source_capture": "https://frontierlog.ai/captures/478a8e3e7a393b2d.png"
          },
          {
            "id": "gpt-6-astra:arc-agi-2:default: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": "arc-agi-2",
            "benchmark_name": "ARC-AGI-2",
            "setting": "default",
            "setting_name": "",
            "score": 95.0,
            "unit": "%",
            "printed": "95.0%",
            "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": "'Abstract reasoning' table, column 'GPT‑6 Astra', row 'ARC-AGI-2'",
            "own": true,
            "primary": true,
            "parallel": false,
            "notes": "",
            "caveat": "",
            "human_verified": true,
            "human_verified_on": "2026-10-17",
            "check": "human",
            "corroborated": false,
            "corroborated_on": null,
            "corroborated_by": [],
            "source_capture": "https://frontierlog.ai/captures/6917e15279360166.png"
          },
          {
            "id": "gpt-6-astra:browsecomp:default: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": "browsecomp",
            "benchmark_name": "BrowseComp",
            "setting": "default",
            "setting_name": "",
            "score": 91.5,
            "unit": "%",
            "printed": "91.5%",
            "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": "'Professional' eval table, row 'BrowseComp', column 'GPT‑6 Astra'",
            "own": true,
            "primary": true,
            "parallel": false,
            "notes": "",
            "caveat": "",
            "human_verified": true,
            "human_verified_on": "2026-10-17",
            "check": "human",
            "corroborated": false,
            "corroborated_on": null,
            "corroborated_by": [],
            "source_capture": "https://frontierlog.ai/captures/8618c81cd990eeed.png"
          },
          {
            "id": "gpt-6-astra:arc-agi-1:default: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": "arc-agi-1",
            "benchmark_name": "ARC-AGI-1",
            "setting": "default",
            "setting_name": "",
            "score": 98.5,
            "unit": "%",
            "printed": "98.5%",
            "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": "'Abstract reasoning' eval table (HTML), row 'ARC-AGI-1', column 'GPT‑6 Astra'",
            "own": true,
            "primary": true,
            "parallel": false,
            "notes": "",
            "caveat": "",
            "human_verified": true,
            "human_verified_on": "2026-10-18",
            "check": "human",
            "corroborated": false,
            "corroborated_on": null,
            "corroborated_by": [],
            "source_capture": "https://frontierlog.ai/captures/6917e15279360166.png"
          },
          {
            "id": "gpt-6-astra:arc-agi-3:default: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": "arc-agi-3",
            "benchmark_name": "ARC-AGI-3",
            "setting": "default",
            "setting_name": "",
            "score": 99.9,
            "unit": "%",
            "printed": "99.9%",
            "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": "'Abstract reasoning' eval table (HTML), row 'ARC-AGI-3', column 'GPT‑6 Astra'",
            "own": true,
            "primary": true,
            "parallel": false,
            "notes": "",
            "caveat": "",
            "human_verified": true,
            "human_verified_on": "2026-10-18",
            "check": "human",
            "corroborated": false,
            "corroborated_on": null,
            "corroborated_by": [],
            "source_capture": "https://frontierlog.ai/captures/6917e15279360166.png"
          },
          {
            "id": "gpt-6-astra:frontiercode-1-1:extended: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": "frontiercode-1-1",
            "benchmark_name": "FrontierCode 1.1",
            "setting": "extended",
            "setting_name": "Extended set",
            "score": 64.5,
            "unit": "%",
            "printed": "64.5%",
            "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": "'Coding' table, row 'FrontierCode 1.1 Extended (score)', column 'GPT-6 Astra' (cell carries footnote 8)",
            "own": true,
            "primary": true,
            "parallel": false,
            "notes": "",
            "caveat": "",
            "human_verified": true,
            "human_verified_on": "2026-10-20",
            "check": "human",
            "corroborated": false,
            "corroborated_on": null,
            "corroborated_by": [],
            "source_capture": "https://frontierlog.ai/captures/4de8f321fbadf8b8.png"
          },
          {
            "id": "gpt-6-astra:gpqa-diamond:default:2",
            "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": "gpqa-diamond",
            "benchmark_name": "GPQA Diamond",
            "setting": "default",
            "setting_name": "",
            "score": 94.9,
            "unit": "%",
            "printed": "94.9%",
            "reported_by": "openai",
            "run_by": null,
            "label": "GPT-6 Astra (lower-cost setting)",
            "source": "https://openai.com/index/gpt-6-astra",
            "source_title": "GPT-6 Astra: A new generation of intelligence",
            "source_date": "2026-09-03",
            "where": "paragraph on GPQA Diamond under the 'Academic' results, the lower-cost-setting figure for Astra",
            "own": true,
            "primary": false,
            "parallel": false,
            "notes": "lower-cost setting; effort not stated",
            "caveat": "",
            "human_verified": true,
            "human_verified_on": "2026-10-20",
            "check": "human",
            "corroborated": false,
            "corroborated_on": null,
            "corroborated_by": []
          },
          {
            "id": "gpt-6-astra:terminal-bench-science-0-1:default:2",
            "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": "terminal-bench-science-0-1",
            "benchmark_name": "Terminal-Bench Science 0.1",
            "setting": "default",
            "setting_name": "",
            "score": 61.1,
            "unit": "%",
            "printed": "61.1%",
            "reported_by": "openai",
            "run_by": null,
            "label": "GPT-6 Astra (lower-cost setting)",
            "source": "https://openai.com/index/gpt-6-astra",
            "source_title": "GPT-6 Astra: A new generation of intelligence",
            "source_date": "2026-09-03",
            "where": "paragraph on Terminal-Bench Science 0.1, sentence 'At a lower-cost setting, Astra scores ...'",
            "own": true,
            "primary": false,
            "parallel": false,
            "notes": "lower-cost setting; effort not stated",
            "caveat": "",
            "human_verified": true,
            "human_verified_on": "2026-10-20",
            "check": "human",
            "corroborated": false,
            "corroborated_on": null,
            "corroborated_by": []
          },
          {
            "id": "gpt-6-astra:automationbench:default:2",
            "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": "automationbench",
            "benchmark_name": "AutomationBench",
            "setting": "default",
            "setting_name": "",
            "score": 30.3,
            "unit": "%",
            "printed": "30.3%",
            "reported_by": "openai",
            "run_by": null,
            "label": "GPT-6 Astra (low)",
            "source": "https://openai.com/index/introducing-gpt-6-sol-and-luna",
            "source_title": "Introducing GPT-6 Sol and Luna",
            "source_date": "2026-09-22",
            "where": "AutomationBench table (Model (and effort) / Score / Cost per task), row 'GPT-6 Astra (low)', column 'Score'",
            "own": false,
            "primary": false,
            "parallel": false,
            "notes": "low effort; AutomationBench 1.0.6",
            "caveat": "",
            "human_verified": true,
            "human_verified_on": "2026-10-20",
            "check": "human",
            "corroborated": false,
            "corroborated_on": null,
            "corroborated_by": [],
            "source_capture": "https://frontierlog.ai/captures/60118f343b9598a5.png"
          },
          {
            "id": "gpt-6-astra:terminal-bench-science-0-1:default:3",
            "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": "terminal-bench-science-0-1",
            "benchmark_name": "Terminal-Bench Science 0.1",
            "setting": "default",
            "setting_name": "",
            "score": 68.1,
            "unit": "%",
            "printed": "68.1%",
            "reported_by": "openai",
            "run_by": null,
            "label": "GPT-6 Astra",
            "source": "https://openai.com/index/introducing-gpt-6-1-sol",
            "source_title": "Introducing GPT-6.1 Sol",
            "source_date": "2026-09-29",
            "where": "'Scientific research' section, paragraph after the Terminal-Bench Science 0.1 cost sentence: 'GPT-6 Astra still achieves the highest score among the models tested at ...'",
            "own": false,
            "primary": false,
            "parallel": false,
            "notes": "maximum effort",
            "caveat": "",
            "human_verified": true,
            "human_verified_on": "2026-10-20",
            "check": "human",
            "corroborated": false,
            "corroborated_on": null,
            "corroborated_by": []
          },
          {
            "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": true,
            "human_verified_on": "2026-10-30",
            "check": "human",
            "corroborated": false,
            "corroborated_on": null,
            "corroborated_by": []
          },
          {
            "id": "gpt-6-astra:sec-bench-pro-no-release:default: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": "sec-bench-pro-no-release",
            "benchmark_name": "SEC-bench Pro, no release named",
            "setting": "default",
            "setting_name": "",
            "score": 85.4,
            "unit": "%",
            "printed": "85.4%",
            "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": "Cybersecurity eval table (HTML; columns GPT‑6 Astra, GPT‑5.6 Sol, Claude Fable 5.1, ...), row 'SEC-Bench Pro', column 'GPT‑6 Astra'",
            "own": true,
            "primary": true,
            "parallel": false,
            "notes": "",
            "caveat": "",
            "human_verified": false,
            "human_verified_on": null,
            "check": "corroborated",
            "corroborated": true,
            "corroborated_on": "2026-10-30",
            "corroborated_by": [
              {
                "url": "https://storage.googleapis.com/deepmind-media/gemini/gemini_4_carbon_model_evaluation.pdf",
                "name": "Gemini 4 Carbon card",
                "title": "Gemini 4 Carbon Model evaluation: approach, methodology & results",
                "lab": "google",
                "date": "2026-10-28",
                "printed": "85.4%",
                "where": "page 5 'Results' table (raster image), row 'SEC-bench Pro', column 'GPT-6 Astra' (Google restating; footnote: competitor numbers taken from their published system cards and technical reports)"
              },
              {
                "url": "https://blog.google/innovation-and-ai/models-and-research/gemini-models/introducing-gemini-4-carbon/",
                "name": "Gemini 4 Carbon",
                "title": "Gemini 4 Carbon: our next frontier model",
                "lab": "google",
                "date": "2026-10-28",
                "printed": "85.4%",
                "where": "SEC-bench Pro bar chart (Pass@1), value label on bar 'GPT-6 Astra'"
              }
            ]
          }
        ],
        "price": [
          {
            "from": "2026-09-03",
            "listed": false,
            "currency": "USD",
            "input_per_1m": 10.0,
            "cached_input_per_1m": 1.0,
            "cached_input_derived": false,
            "output_per_1m": 50.0,
            "cache": {
              "kind": "price"
            },
            "tiers": [
              {
                "over": 272000,
                "measure": "input",
                "input_per_1m": 20.0,
                "cached_input_per_1m": 2.0,
                "cached_input_derived": false,
                "output_per_1m": 75.0,
                "printed": {
                  "input": "$20.00",
                  "cached_input": "$2.00",
                  "output": "$75.00"
                }
              }
            ],
            "printed": {
              "input": "$10",
              "cached_input": "$1.00",
              "output": "$50"
            },
            "printed_unit": "1M tokens",
            "notes": "",
            "sources": [
              "https://openai.com/index/gpt-6-astra",
              "https://openai.com/index/introducing-gpt-6-1-sol",
              "https://developers.openai.com/api/docs/pricing",
              "https://developers.openai.com/api/docs/models/gpt-6-astra.md",
              "https://developers.openai.com/api/docs/models",
              "https://developers.openai.com/api/docs/models/gpt-6-astra"
            ],
            "check": "corroborated"
          }
        ],
        "context": {
          "from": "2026-11-04",
          "listed": true,
          "window_tokens": 1050000,
          "input_tokens": 922000,
          "output_tokens": 128000,
          "tiers": [],
          "printed": {
            "window": "1,050,000",
            "input": "922,000",
            "output": "128,000"
          },
          "notes": "",
          "sources": [
            "https://developers.openai.com/api/docs/models/gpt-6-astra.md",
            "https://developers.openai.com/api/docs/models/compare?model=gpt-6-astra",
            "https://developers.openai.com/api/docs/models"
          ],
          "check": "corroborated"
        },
        "prices": [
          {
            "from": "2026-09-03",
            "listed": false,
            "currency": "USD",
            "input_per_1m": 10.0,
            "cached_input_per_1m": 1.0,
            "cached_input_derived": false,
            "output_per_1m": 50.0,
            "cache": {
              "kind": "price"
            },
            "tiers": [
              {
                "over": 272000,
                "measure": "input",
                "input_per_1m": 20.0,
                "cached_input_per_1m": 2.0,
                "cached_input_derived": false,
                "output_per_1m": 75.0,
                "printed": {
                  "input": "$20.00",
                  "cached_input": "$2.00",
                  "output": "$75.00"
                }
              }
            ],
            "printed": {
              "input": "$10",
              "cached_input": "$1.00",
              "output": "$50"
            },
            "printed_unit": "1M tokens",
            "notes": "",
            "sources": [
              "https://openai.com/index/gpt-6-astra",
              "https://openai.com/index/introducing-gpt-6-1-sol",
              "https://developers.openai.com/api/docs/pricing",
              "https://developers.openai.com/api/docs/models/gpt-6-astra.md",
              "https://developers.openai.com/api/docs/models",
              "https://developers.openai.com/api/docs/models/gpt-6-astra"
            ],
            "check": "corroborated"
          }
        ],
        "context_history": [
          {
            "from": "2026-11-04",
            "listed": true,
            "window_tokens": 1050000,
            "input_tokens": 922000,
            "output_tokens": 128000,
            "tiers": [],
            "printed": {
              "window": "1,050,000",
              "input": "922,000",
              "output": "128,000"
            },
            "notes": "",
            "sources": [
              "https://developers.openai.com/api/docs/models/gpt-6-astra.md",
              "https://developers.openai.com/api/docs/models/compare?model=gpt-6-astra",
              "https://developers.openai.com/api/docs/models"
            ],
            "check": "corroborated"
          }
        ],
        "stories": [
          {
            "slug": "chatgpt-plan-gpt-6-models",
            "title": "In ChatGPT, GPT-6 Is Sol and GPT-6 Pro Is Astra",
            "date": "2026-11-05",
            "url": "https://frontierlog.ai/stories/chatgpt-plan-gpt-6-models"
          },
          {
            "slug": "meta-muse-rise",
            "title": "Meta's Muse Models Climbed to the Top of DeepSWE in Six Months",
            "date": "2026-11-04",
            "url": "https://frontierlog.ai/stories/meta-muse-rise"
          },
          {
            "slug": "gpt-6-family",
            "title": "The GPT-6 Family Tree: Astra, Sol, Terra and Luna",
            "date": "2026-11-04",
            "url": "https://frontierlog.ai/stories/gpt-6-family"
          },
          {
            "slug": "who-leads-terminal-bench-4",
            "title": "Who Leads Terminal-Bench 4.0 Depends on Who Ran It",
            "date": "2026-10-30",
            "url": "https://frontierlog.ai/stories/who-leads-terminal-bench-4"
          },
          {
            "slug": "open-models-frontiercode-harness",
            "title": "FrontierCode 2.0's Top Open Model Depends on the Harness",
            "date": "2026-10-30",
            "url": "https://frontierlog.ai/stories/open-models-frontiercode-harness"
          },
          {
            "slug": "gemini-4-carbon-aa-scores",
            "title": "Gemini 4 Carbon's Table Moves Rivals' Scores",
            "date": "2026-10-29",
            "url": "https://frontierlog.ai/stories/gemini-4-carbon-aa-scores"
          },
          {
            "slug": "openai-training-pause",
            "title": "OpenAI Paused Frontier Training for 11 Days",
            "date": "2026-10-25",
            "url": "https://frontierlog.ai/stories/openai-training-pause"
          },
          {
            "slug": "openai-agents-weakened-monitoring",
            "title": "Two OpenAI Agents Weakened Their Own Monitoring",
            "date": "2026-10-25",
            "url": "https://frontierlog.ai/stories/openai-agents-weakened-monitoring"
          },
          {
            "slug": "china-leads-on-automationbench",
            "title": "China Leads on AutomationBench",
            "date": "2026-10-24",
            "url": "https://frontierlog.ai/stories/china-leads-on-automationbench"
          },
          {
            "slug": "benchmark-scores-move-after-launch-day",
            "title": "Benchmark Scores Move After Launch Day",
            "date": "2026-10-22",
            "url": "https://frontierlog.ai/stories/benchmark-scores-move-after-launch-day"
          },
          {
            "slug": "benchmark-churn",
            "title": "Which Benchmarks AI Labs Report at Launch",
            "date": "2026-10-18",
            "url": "https://frontierlog.ai/stories/benchmark-churn"
          },
          {
            "slug": "china-vs-the-rest",
            "title": "Where China's Open Models Lead and Trail",
            "date": "2026-10-17",
            "url": "https://frontierlog.ai/stories/china-vs-the-rest"
          }
        ]
      }
    ],
    "source_title": "Covert underperformance on safety research tasks",
    "source_archived": null,
    "source_http_status": null,
    "source_quote": "Astra completed about 10 percent of hidden rubric items under the safety framing, compared with about 16 percent under the capabilities framing.",
    "followups": [],
    "successor": null,
    "predecessors": [],
    "revealed_as": null,
    "stealth_previews": [],
    "reveal_source": null,
    "url": "https://frontierlog.ai/event/2026-11-04-openai-reports-astra-underperformance",
    "image": "https://frontierlog.ai/og/2026-11-04-openai-reports-astra-underperformance.png",
    "stories": []
  }
}