{
  "api_version": "1",
  "updated": "2026-10-26",
  "latest_event": "2026-10-24",
  "rights": "© Frontier Log, all rights reserved.",
  "model": {
    "slug": "gemini-3-pro",
    "name": "Gemini 3 Pro",
    "lab": "google",
    "date": "2025-11-18",
    "flagship": true,
    "event": "https://frontierlog.ai/event/2025-11-18-gemini-3-pro",
    "releases": [],
    "variant_of": null,
    "made_by": "Google",
    "url": "https://frontierlog.ai/model/gemini-3-pro",
    "results": [
      {
        "id": "gemini-3-pro:hle:no-tools:1",
        "model": "gemini-3-pro",
        "model_name": "Gemini 3 Pro",
        "model_date": "2025-11-18",
        "event": "https://frontierlog.ai/event/2025-11-18-gemini-3-pro",
        "benchmark": "hle",
        "benchmark_name": "Humanity's Last Exam",
        "setting": "no-tools",
        "setting_name": "No tools",
        "score": 37.5,
        "unit": "%",
        "printed": "37.5%",
        "reported_by": "google",
        "run_by": null,
        "label": "Gemini 3 Pro",
        "source": "https://blog.google/innovation-and-ai/technology/developers-tools/gemini-3-developers/",
        "source_title": "Start building with Gemini 3",
        "source_date": "2025-11-18",
        "where": "benchmark table (image), row 'Humanity's Last Exam', 'No tools', column 'Gemini 3 Pro'",
        "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/news/claude-opus-4-6",
            "name": "Claude Opus 4.6",
            "title": "Claude Opus 4.6",
            "lab": "anthropic",
            "date": "2026-02-05",
            "printed": "37.5%",
            "where": "Benchmark table image, row 'Multidisciplinary reasoning Humanity's Last Exam, 'without tools'', column 'Gemini 3 Pro'"
          },
          {
            "url": "https://blog.google/innovation-and-ai/models-and-research/gemini-models/gemini-3-1-pro/",
            "name": "Gemini 3.1 Pro",
            "title": "Gemini 3.1 Pro: A smarter model for your most complex tasks",
            "lab": "google",
            "date": "2026-02-19",
            "printed": "37.5%",
            "where": "Benchmark table image; row 'Humanity's Last Exam, No tools'; column gemini-3-pro"
          },
          {
            "url": "https://qwen.ai/blog?id=qwen3-max-thinking",
            "name": "Qwen3-Max-Thinking",
            "title": "Pushing Qwen3-Max-Thinking Beyond its Limits",
            "lab": "alibaba",
            "date": "2026-01-26",
            "printed": "37.5",
            "where": "Score chart image (score.png), panel Humanity's Last Exam, value label on the Gemini-3 Pro bar"
          },
          {
            "url": "https://z.ai/blog/glm-4.7",
            "name": "GLM-4.7",
            "title": "GLM-4.7: Advancing the Coding Capability",
            "lab": "zai",
            "date": "2025-12-22",
            "printed": "37.5",
            "where": "Benchmark table, row 'HLE (separate from 'HLE w/ Tools' row)', column 'Gemini 3.0 Pro'"
          }
        ],
        "source_image": "https://frontierlog.ai/sources/3fd7d3190ca58049.gif",
        "source_image_original": "https://storage.googleapis.com/gweb-uniblog-publish-prod/documents/gemini_3_table_final_HLE_Tools_on.gif"
      },
      {
        "id": "gemini-3-pro:arc-agi-2:default:1",
        "model": "gemini-3-pro",
        "model_name": "Gemini 3 Pro",
        "model_date": "2025-11-18",
        "event": "https://frontierlog.ai/event/2025-11-18-gemini-3-pro",
        "benchmark": "arc-agi-2",
        "benchmark_name": "ARC-AGI-2",
        "setting": "default",
        "setting_name": "",
        "score": 31.1,
        "unit": "%",
        "printed": "31.1%",
        "reported_by": "google",
        "run_by": null,
        "label": "Gemini 3 Pro",
        "source": "https://blog.google/innovation-and-ai/technology/developers-tools/gemini-3-developers/",
        "source_title": "Start building with Gemini 3",
        "source_date": "2025-11-18",
        "where": "benchmark table (image), row 'ARC-AGI-2', 'ARC Prize Verified', column 'Gemini 3 Pro'",
        "own": true,
        "primary": true,
        "parallel": false,
        "notes": "ARC Prize Verified",
        "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-4-5-system-card",
            "name": "Claude Opus 4.5 card",
            "title": "Claude Opus 4.5 System Card",
            "lab": "anthropic",
            "date": "2025-11-24",
            "printed": "31.1%",
            "where": "Table 2.3.A (p.19), row ARC-AGI-2 (Verified), column Gemini 3 Pro"
          },
          {
            "url": "https://www.anthropic.com/news/claude-opus-4-5",
            "name": "Claude Opus 4.5",
            "title": "Introducing Claude Opus 4.5",
            "lab": "anthropic",
            "date": "2025-11-24",
            "printed": "31.1%",
            "where": "Benchmark table image, row 'Novel problem solving ARC-AGI-2 (Verified)', column 'Gemini 3 Pro'"
          },
          {
            "url": "https://blog.google/innovation-and-ai/models-and-research/gemini-models/gemini-3-1-pro/",
            "name": "Gemini 3.1 Pro",
            "title": "Gemini 3.1 Pro: A smarter model for your most complex tasks",
            "lab": "google",
            "date": "2026-02-19",
            "printed": "31.1%",
            "where": "Benchmark table image; row 'ARC-AGI-2, ARC Prize Verified'; column gemini-3-pro"
          }
        ],
        "source_image": "https://frontierlog.ai/sources/3fd7d3190ca58049.gif",
        "source_image_original": "https://storage.googleapis.com/gweb-uniblog-publish-prod/documents/gemini_3_table_final_HLE_Tools_on.gif"
      },
      {
        "id": "gemini-3-pro:gpqa-diamond:default:1",
        "model": "gemini-3-pro",
        "model_name": "Gemini 3 Pro",
        "model_date": "2025-11-18",
        "event": "https://frontierlog.ai/event/2025-11-18-gemini-3-pro",
        "benchmark": "gpqa-diamond",
        "benchmark_name": "GPQA Diamond",
        "setting": "default",
        "setting_name": "",
        "score": 91.9,
        "unit": "%",
        "printed": "91.9%",
        "reported_by": "google",
        "run_by": null,
        "label": "Gemini 3 Pro",
        "source": "https://blog.google/innovation-and-ai/technology/developers-tools/gemini-3-developers/",
        "source_title": "Start building with Gemini 3",
        "source_date": "2025-11-18",
        "where": "benchmark table (image), row 'GPQA Diamond', 'No tools', column 'Gemini 3 Pro'",
        "own": true,
        "primary": true,
        "parallel": false,
        "notes": "no tools",
        "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-4-5-system-card",
            "name": "Claude Opus 4.5 card",
            "title": "Claude Opus 4.5 System Card",
            "lab": "anthropic",
            "date": "2025-11-24",
            "printed": "91.9%",
            "where": "Table 2.3.A (p.19), row GPQA Diamond, column Gemini 3 Pro"
          },
          {
            "url": "https://www.anthropic.com/claude-opus-4-6-system-card",
            "name": "Claude Opus 4.6 card",
            "title": "Claude Opus 4.6 System Card",
            "lab": "anthropic",
            "date": "2026-02-05",
            "printed": "91.9%",
            "where": "Table 2.3.A (p.18), row GPQA Diamond, column Gemini 3 Pro"
          },
          {
            "url": "https://www.anthropic.com/news/claude-opus-4-5",
            "name": "Claude Opus 4.5",
            "title": "Introducing Claude Opus 4.5",
            "lab": "anthropic",
            "date": "2025-11-24",
            "printed": "91.9%",
            "where": "Benchmark table image, row 'Graduate-level reasoning GPQA Diamond', column 'Gemini 3 Pro'"
          },
          {
            "url": "https://www.anthropic.com/news/claude-opus-4-6",
            "name": "Claude Opus 4.6",
            "title": "Claude Opus 4.6",
            "lab": "anthropic",
            "date": "2026-02-05",
            "printed": "91.9%",
            "where": "Benchmark table image, row 'Graduate-level reasoning GPQA Diamond', column 'Gemini 3 Pro'"
          },
          {
            "url": "https://api-docs.deepseek.com/news/news251201/",
            "name": "DeepSeek-V3.2",
            "title": "DeepSeek-V3.2 Release",
            "lab": "deepseek",
            "date": "2025-12-01",
            "printed": "91.9",
            "where": "Reasoning table image (token counts in parentheses omitted); row 'GPQA Diamond'; column 'Gemini-3.0 Pro'"
          },
          {
            "url": "https://blog.google/innovation-and-ai/models-and-research/gemini-models/gemini-3-1-pro/",
            "name": "Gemini 3.1 Pro",
            "title": "Gemini 3.1 Pro: A smarter model for your most complex tasks",
            "lab": "google",
            "date": "2026-02-19",
            "printed": "91.9%",
            "where": "Benchmark table image; row 'GPQA Diamond, No tools'; column gemini-3-pro"
          },
          {
            "url": "https://www.kimi.ai/blog/kimi-k2-5",
            "name": "Kimi K2.5",
            "title": "Kimi K2.5: Visual Agentic Intelligence",
            "lab": "moonshot",
            "date": "2026-01-27",
            "printed": "91.9",
            "where": "Appendix Benchmark table, row GPQA-Diamond (avg@8) ; column Gemini 3 Pro (High Thinking Level)"
          },
          {
            "url": "https://qwen.ai/blog?id=qwen3-max-thinking",
            "name": "Qwen3-Max-Thinking",
            "title": "Pushing Qwen3-Max-Thinking Beyond its Limits",
            "lab": "alibaba",
            "date": "2026-01-26",
            "printed": "91.9",
            "where": "Score chart image (score.png), panel GPQA Diamond, value label on the Gemini-3 Pro bar"
          },
          {
            "url": "https://z.ai/blog/glm-4.7",
            "name": "GLM-4.7",
            "title": "GLM-4.7: Advancing the Coding Capability",
            "lab": "zai",
            "date": "2025-12-22",
            "printed": "91.9",
            "where": "Benchmark table, row 'GPQA-Diamond', column 'Gemini 3.0 Pro'"
          },
          {
            "url": "https://z.ai/blog/glm-5",
            "name": "GLM-5",
            "title": "GLM-5: From Vibe Coding to Agentic Engineering",
            "lab": "zai",
            "date": "2026-02-12",
            "printed": "91.9",
            "where": "Benchmark table, row 'GPQA-Diamond', column 'Gemini 3.0 Pro (High Thinking Level)'"
          }
        ],
        "source_image": "https://frontierlog.ai/sources/3fd7d3190ca58049.gif",
        "source_image_original": "https://storage.googleapis.com/gweb-uniblog-publish-prod/documents/gemini_3_table_final_HLE_Tools_on.gif"
      },
      {
        "id": "gemini-3-pro:terminal-bench-2-0:default:1",
        "model": "gemini-3-pro",
        "model_name": "Gemini 3 Pro",
        "model_date": "2025-11-18",
        "event": "https://frontierlog.ai/event/2025-11-18-gemini-3-pro",
        "benchmark": "terminal-bench-2-0",
        "benchmark_name": "Terminal-Bench 2.0",
        "setting": "default",
        "setting_name": "",
        "score": 54.2,
        "unit": "%",
        "printed": "54.2%",
        "reported_by": "google",
        "run_by": null,
        "label": "Gemini 3 Pro",
        "source": "https://blog.google/innovation-and-ai/technology/developers-tools/gemini-3-developers/",
        "source_title": "Start building with Gemini 3",
        "source_date": "2025-11-18",
        "where": "benchmark table (image), row 'Terminal-Bench 2.0', 'Terminus-2 agent', column 'Gemini 3 Pro'",
        "own": true,
        "primary": true,
        "parallel": false,
        "notes": "Terminus 2 harness",
        "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-4-5-system-card",
            "name": "Claude Opus 4.5 card",
            "title": "Claude Opus 4.5 System Card",
            "lab": "anthropic",
            "date": "2025-11-24",
            "printed": "54.2%",
            "where": "Table 2.3.A (p.19), row Terminal-bench 2.0, column Gemini 3 Pro"
          },
          {
            "url": "https://www.anthropic.com/news/claude-opus-4-5",
            "name": "Claude Opus 4.5",
            "title": "Introducing Claude Opus 4.5",
            "lab": "anthropic",
            "date": "2025-11-24",
            "printed": "54.2%",
            "where": "Benchmark table image, row 'Agentic terminal coding Terminal-bench 2.0', column 'Gemini 3 Pro'"
          },
          {
            "url": "https://www.anthropic.com/news/claude-opus-4-6",
            "name": "Claude Opus 4.6",
            "title": "Claude Opus 4.6",
            "lab": "anthropic",
            "date": "2026-02-05",
            "printed": "54.2%",
            "where": "Benchmark table image, row 'Agentic terminal coding Terminal-Bench 2.0', column 'Gemini 3 Pro', small text '(54.2% self-reported)'"
          },
          {
            "url": "https://api-docs.deepseek.com/news/news251201/",
            "name": "DeepSeek-V3.2",
            "title": "DeepSeek-V3.2 Release",
            "lab": "deepseek",
            "date": "2025-12-01",
            "printed": "54.2",
            "where": "Bar chart image; group 'Terminal Bench 2.0 (Acc)'; value label on Gemini-3.0-Pro bar"
          },
          {
            "url": "https://www.minimax.io/news/minimax-m21",
            "name": "MiniMax M2.1",
            "title": "MiniMax M2.1: Significantly Enhanced Multi-Language Programming, Built for Real-World Complex Tasks",
            "lab": "minimax",
            "date": "2025-12-23",
            "printed": "54.2",
            "where": "Coding bar chart image (10 panels); panel 'Terminal-bench 2.0'; value label on Gemini 3 Pro bar"
          },
          {
            "url": "https://www.kimi.ai/blog/kimi-k2-5",
            "name": "Kimi K2.5",
            "title": "Kimi K2.5: Visual Agentic Intelligence",
            "lab": "moonshot",
            "date": "2026-01-27",
            "printed": "54.2",
            "where": "Appendix Benchmark table, row Terminal-Bench 2.0 (Terminus-2; K2.5 non-thinking mode) ; column Gemini 3 Pro (High Thinking Level)"
          },
          {
            "url": "https://qwen.ai/blog?id=qwen3.5",
            "name": "Qwen3.5",
            "title": "Qwen3.5: Towards Native Multimodal Agents",
            "lab": "alibaba",
            "date": "2026-02-16",
            "printed": "54.2",
            "where": "Score chart image (qwen3.5_397b_a17b_score.png), panel Terminal-Bench 2, value label on the Gemini 3 Pro bar"
          },
          {
            "url": "https://z.ai/blog/glm-4.7",
            "name": "GLM-4.7",
            "title": "GLM-4.7: Advancing the Coding Capability",
            "lab": "zai",
            "date": "2025-12-22",
            "printed": "54.2",
            "where": "Benchmark table, row 'Terminal Bench 2.0', column 'Gemini 3.0 Pro'"
          },
          {
            "url": "https://z.ai/blog/glm-5",
            "name": "GLM-5",
            "title": "GLM-5: From Vibe Coding to Agentic Engineering",
            "lab": "zai",
            "date": "2026-02-12",
            "printed": "54.2",
            "where": "Benchmark table, row 'Terminal-Bench 2.0 Terminus-2 (GLM-5 cell '56.2 / 60.7†': first value; 60.7 is the verified variant)', column 'Gemini 3.0 Pro (High Thinking Level)'"
          }
        ],
        "source_image": "https://frontierlog.ai/sources/3fd7d3190ca58049.gif",
        "source_image_original": "https://storage.googleapis.com/gweb-uniblog-publish-prod/documents/gemini_3_table_final_HLE_Tools_on.gif"
      },
      {
        "id": "gemini-3-pro:terminal-bench-2-0:default:2",
        "model": "gemini-3-pro",
        "model_name": "Gemini 3 Pro",
        "model_date": "2025-11-18",
        "event": "https://frontierlog.ai/event/2025-11-18-gemini-3-pro",
        "benchmark": "terminal-bench-2-0",
        "benchmark_name": "Terminal-Bench 2.0",
        "setting": "default",
        "setting_name": "",
        "score": 56.9,
        "unit": "%",
        "printed": "56.9%",
        "reported_by": "google",
        "run_by": null,
        "label": "Gemini 3 Pro, Thinking (High)",
        "source": "https://blog.google/innovation-and-ai/models-and-research/gemini-models/gemini-3-1-pro/",
        "source_title": "Gemini 3.1 Pro: A smarter model for your most complex tasks",
        "source_date": "2026-02-19",
        "where": "benchmark table (image), row 'Terminal-Bench 2.0', 'Terminus-2 harness', column 'Gemini 3 Pro, Thinking (High)'",
        "own": false,
        "primary": false,
        "parallel": false,
        "notes": "Terminus 2 harness; Thinking (High)",
        "caveat": "",
        "human_verified": true,
        "human_verified_on": "2026-10-16",
        "check": "human",
        "corroborated": false,
        "corroborated_on": null,
        "corroborated_by": [],
        "source_image": "https://frontierlog.ai/sources/6b612deebae92383.gif",
        "source_image_original": "https://storage.googleapis.com/gweb-uniblog-publish-prod/original_images/gemini_3-1-pro__benchmarks.gif"
      },
      {
        "id": "gemini-3-pro:swe-bench-verified:default:1",
        "model": "gemini-3-pro",
        "model_name": "Gemini 3 Pro",
        "model_date": "2025-11-18",
        "event": "https://frontierlog.ai/event/2025-11-18-gemini-3-pro",
        "benchmark": "swe-bench-verified",
        "benchmark_name": "SWE-bench Verified",
        "setting": "default",
        "setting_name": "",
        "score": 76.2,
        "unit": "%",
        "printed": "76.2%",
        "reported_by": "google",
        "run_by": null,
        "label": "Gemini 3 Pro",
        "source": "https://blog.google/innovation-and-ai/technology/developers-tools/gemini-3-developers/",
        "source_title": "Start building with Gemini 3",
        "source_date": "2025-11-18",
        "where": "benchmark table (image), row 'SWE-Bench Verified', 'Single attempt', column 'Gemini 3 Pro'",
        "own": true,
        "primary": true,
        "parallel": false,
        "notes": "single attempt",
        "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-4-5-system-card",
            "name": "Claude Opus 4.5 card",
            "title": "Claude Opus 4.5 System Card",
            "lab": "anthropic",
            "date": "2025-11-24",
            "printed": "76.2%",
            "where": "Table 2.3.A (p.19), row SWE-bench Verified, column Gemini 3 Pro"
          },
          {
            "url": "https://www.anthropic.com/claude-opus-4-6-system-card",
            "name": "Claude Opus 4.6 card",
            "title": "Claude Opus 4.6 System Card",
            "lab": "anthropic",
            "date": "2026-02-05",
            "printed": "76.2%",
            "where": "Table 2.3.A (p.18), row SWE-bench Verified, column Gemini 3 Pro"
          },
          {
            "url": "https://www.anthropic.com/news/claude-opus-4-5",
            "name": "Claude Opus 4.5",
            "title": "Introducing Claude Opus 4.5",
            "lab": "anthropic",
            "date": "2025-11-24",
            "printed": "76.2%",
            "where": "Benchmark table image, row 'Agentic coding SWE-bench Verified', column 'Gemini 3 Pro'"
          },
          {
            "url": "https://www.anthropic.com/news/claude-opus-4-6",
            "name": "Claude Opus 4.6",
            "title": "Claude Opus 4.6",
            "lab": "anthropic",
            "date": "2026-02-05",
            "printed": "76.2%",
            "where": "Benchmark table image, row 'Agentic coding SWE-bench Verified', column 'Gemini 3 Pro'"
          },
          {
            "url": "https://api-docs.deepseek.com/news/news251201/",
            "name": "DeepSeek-V3.2",
            "title": "DeepSeek-V3.2 Release",
            "lab": "deepseek",
            "date": "2025-12-01",
            "printed": "76.2",
            "where": "Bar chart image; group 'SWE Verified (Resolved)'; value label on Gemini-3.0-Pro bar"
          },
          {
            "url": "https://www.minimax.io/news/minimax-m25",
            "name": "MiniMax M2.5",
            "title": "MiniMax M2.5: Built for Real-World Productivity.",
            "lab": "minimax",
            "date": "2026-02-12",
            "printed": "76.2%",
            "where": "'SWE-bench Verified Score Evolution' line chart image; point label 'Gemini 3.0 Pro 76.2%'"
          },
          {
            "url": "https://blog.google/innovation-and-ai/models-and-research/gemini-models/gemini-3-1-pro/",
            "name": "Gemini 3.1 Pro",
            "title": "Gemini 3.1 Pro: A smarter model for your most complex tasks",
            "lab": "google",
            "date": "2026-02-19",
            "printed": "76.2%",
            "where": "Benchmark table image; row 'SWE-Bench Verified, Single attempt'; column gemini-3-pro"
          },
          {
            "url": "https://www.kimi.ai/blog/kimi-k2-5",
            "name": "Kimi K2.5",
            "title": "Kimi K2.5: Visual Agentic Intelligence",
            "lab": "moonshot",
            "date": "2026-01-27",
            "printed": "76.2",
            "where": "Appendix Benchmark table, row SWE-Bench Verified ; column Gemini 3 Pro (High Thinking Level)"
          },
          {
            "url": "https://qwen.ai/blog?id=qwen3-max-thinking",
            "name": "Qwen3-Max-Thinking",
            "title": "Pushing Qwen3-Max-Thinking Beyond its Limits",
            "lab": "alibaba",
            "date": "2026-01-26",
            "printed": "76.2",
            "where": "Score chart image (score.png), panel SWE-bench Verified, value label on the Gemini-3 Pro bar"
          },
          {
            "url": "https://mistral.ai/news/devstral-2-vibe-cli/",
            "name": "Devstral 2",
            "title": "Introducing: Devstral 2 and Mistral Vibe CLI.",
            "lab": "mistral",
            "date": "2025-12-09",
            "printed": "76.2",
            "where": "Chart 'SWE-Bench Verified: Open-weight vs Proprietary models', bar 'Gemini 3 Pro'"
          },
          {
            "url": "https://z.ai/blog/glm-4.7",
            "name": "GLM-4.7",
            "title": "GLM-4.7: Advancing the Coding Capability",
            "lab": "zai",
            "date": "2025-12-22",
            "printed": "76.2",
            "where": "Benchmark table, row 'SWE-bench Verified', column 'Gemini 3.0 Pro'"
          },
          {
            "url": "https://z.ai/blog/glm-5",
            "name": "GLM-5",
            "title": "GLM-5: From Vibe Coding to Agentic Engineering",
            "lab": "zai",
            "date": "2026-02-12",
            "printed": "76.2",
            "where": "Benchmark table, row 'SWE-bench Verified', column 'Gemini 3.0 Pro (High Thinking Level)'"
          }
        ],
        "source_image": "https://frontierlog.ai/sources/3fd7d3190ca58049.gif",
        "source_image_original": "https://storage.googleapis.com/gweb-uniblog-publish-prod/documents/gemini_3_table_final_HLE_Tools_on.gif"
      },
      {
        "id": "gemini-3-pro:hle:with-tools:1",
        "model": "gemini-3-pro",
        "model_name": "Gemini 3 Pro",
        "model_date": "2025-11-18",
        "event": "https://frontierlog.ai/event/2025-11-18-gemini-3-pro",
        "benchmark": "hle",
        "benchmark_name": "Humanity's Last Exam",
        "setting": "with-tools",
        "setting_name": "With tools",
        "score": 45.8,
        "unit": "%",
        "printed": "45.8%",
        "reported_by": "google",
        "run_by": null,
        "label": "Gemini 3 Pro",
        "source": "https://blog.google/innovation-and-ai/technology/developers-tools/gemini-3-developers/",
        "source_title": "Start building with Gemini 3",
        "source_date": "2025-11-18",
        "where": "benchmark table (image), row 'Humanity's Last Exam', 'With search and code execution', column 'Gemini 3 Pro'",
        "own": true,
        "primary": true,
        "parallel": false,
        "notes": "search and code execution",
        "caveat": "",
        "human_verified": true,
        "human_verified_on": "2026-10-17",
        "check": "human",
        "corroborated": true,
        "corroborated_on": "2026-10-19",
        "corroborated_by": [
          {
            "url": "https://www.anthropic.com/news/claude-opus-4-6",
            "name": "Claude Opus 4.6",
            "title": "Claude Opus 4.6",
            "lab": "anthropic",
            "date": "2026-02-05",
            "printed": "45.8%",
            "where": "Benchmark table image, row 'Multidisciplinary reasoning Humanity's Last Exam, 'with tools'', column 'Gemini 3 Pro'"
          },
          {
            "url": "https://blog.google/innovation-and-ai/models-and-research/gemini-models/gemini-3-1-pro/",
            "name": "Gemini 3.1 Pro",
            "title": "Gemini 3.1 Pro: A smarter model for your most complex tasks",
            "lab": "google",
            "date": "2026-02-19",
            "printed": "45.8%",
            "where": "Benchmark table image; row 'Humanity's Last Exam, Search (blocklist) + Code'; column gemini-3-pro"
          },
          {
            "url": "https://www.kimi.ai/blog/kimi-k2-5",
            "name": "Kimi K2.5",
            "title": "Kimi K2.5: Visual Agentic Intelligence",
            "lab": "moonshot",
            "date": "2026-01-27",
            "printed": "45.8",
            "where": "Appendix Benchmark table, row HLE-Full w/ tools ; column Gemini 3 Pro (High Thinking Level)"
          },
          {
            "url": "https://z.ai/blog/glm-4.7",
            "name": "GLM-4.7",
            "title": "GLM-4.7: Advancing the Coding Capability",
            "lab": "zai",
            "date": "2025-12-22",
            "printed": "45.8",
            "where": "Benchmark table, row 'HLE w/ Tools', column 'Gemini 3.0 Pro'"
          },
          {
            "url": "https://z.ai/blog/glm-5",
            "name": "GLM-5",
            "title": "GLM-5: From Vibe Coding to Agentic Engineering",
            "lab": "zai",
            "date": "2026-02-12",
            "printed": "45.8",
            "where": "Benchmark table, row 'Humanity's Last Exam w/ Tools (* = full set for Opus 4.5, Gemini, GPT-5.2)', column 'Gemini 3.0 Pro (High Thinking Level)'"
          }
        ],
        "source_image": "https://frontierlog.ai/sources/3fd7d3190ca58049.gif",
        "source_image_original": "https://storage.googleapis.com/gweb-uniblog-publish-prod/documents/gemini_3_table_final_HLE_Tools_on.gif"
      },
      {
        "id": "gemini-3-pro:browsecomp:default:1",
        "model": "gemini-3-pro",
        "model_name": "Gemini 3 Pro",
        "model_date": "2025-11-18",
        "event": "https://frontierlog.ai/event/2025-11-18-gemini-3-pro",
        "benchmark": "browsecomp",
        "benchmark_name": "BrowseComp",
        "setting": "default",
        "setting_name": "",
        "score": 59.2,
        "unit": "%",
        "printed": "59.2%",
        "reported_by": "google",
        "run_by": null,
        "label": "Gemini 3 Pro",
        "source": "https://blog.google/innovation-and-ai/models-and-research/gemini-models/gemini-3-1-pro/",
        "source_title": "Gemini 3.1 Pro: A smarter model for your most complex tasks",
        "source_date": "2026-02-19",
        "where": "benchmark table (image), row 'BrowseComp' (Search + Python + Browse), column 'Gemini 3 Pro Thinking (High)'",
        "own": false,
        "primary": true,
        "parallel": false,
        "notes": "",
        "caveat": "",
        "human_verified": true,
        "human_verified_on": "2026-10-17",
        "check": "human",
        "corroborated": true,
        "corroborated_on": "2026-10-19",
        "corroborated_by": [
          {
            "url": "https://www.minimax.io/news/minimax-m21",
            "name": "MiniMax M2.1",
            "title": "MiniMax M2.1: Significantly Enhanced Multi-Language Programming, Built for Real-World Complex Tasks",
            "lab": "minimax",
            "date": "2025-12-23",
            "printed": "59.2",
            "where": "Tool-use and intelligence table image; row 'BrowseComp (context management)'; column 'Gemini 3 Pro'; with context management"
          },
          {
            "url": "https://www.minimax.io/news/minimax-m25",
            "name": "MiniMax M2.5",
            "title": "MiniMax M2.5: Built for Real-World Productivity.",
            "lab": "minimax",
            "date": "2026-02-12",
            "printed": "59.2",
            "where": "Overview bar chart image (8 panels); panel 'BrowseComp (w/ctx)'; value label on Gemini 3 Pro bar"
          },
          {
            "url": "https://www.kimi.ai/blog/kimi-k2-5",
            "name": "Kimi K2.5",
            "title": "Kimi K2.5: Visual Agentic Intelligence",
            "lab": "moonshot",
            "date": "2026-01-27",
            "printed": "59.2",
            "where": "Appendix Benchmark table, row BrowseComp (w/ctx mgm) ; column Gemini 3 Pro (High Thinking Level)"
          },
          {
            "url": "https://qwen.ai/blog?id=qwen3.5",
            "name": "Qwen3.5",
            "title": "Qwen3.5: Towards Native Multimodal Agents",
            "lab": "alibaba",
            "date": "2026-02-16",
            "printed": "59.2",
            "where": "Score chart image (qwen3.5_397b_a17b_score.png), panel BrowseComp, value label on the Gemini 3 Pro bar"
          },
          {
            "url": "https://z.ai/blog/glm-4.7",
            "name": "GLM-4.7",
            "title": "GLM-4.7: Advancing the Coding Capability",
            "lab": "zai",
            "date": "2025-12-22",
            "printed": "59.2",
            "where": "Benchmark table, row 'BrowseComp w/ Context Manage', column 'Gemini 3.0 Pro'"
          },
          {
            "url": "https://z.ai/blog/glm-5",
            "name": "GLM-5",
            "title": "GLM-5: From Vibe Coding to Agentic Engineering",
            "lab": "zai",
            "date": "2026-02-12",
            "printed": "59.2",
            "where": "Benchmark table, row 'BrowseComp w/ Context Manage', column 'Gemini 3.0 Pro (High Thinking Level)'"
          }
        ],
        "source_image": "https://frontierlog.ai/sources/6b612deebae92383.gif",
        "source_image_original": "https://storage.googleapis.com/gweb-uniblog-publish-prod/original_images/gemini_3-1-pro__benchmarks.gif"
      },
      {
        "id": "gemini-3-pro:swe-bench-pro:default:1",
        "model": "gemini-3-pro",
        "model_name": "Gemini 3 Pro",
        "model_date": "2025-11-18",
        "event": "https://frontierlog.ai/event/2025-11-18-gemini-3-pro",
        "benchmark": "swe-bench-pro",
        "benchmark_name": "SWE-bench Pro",
        "setting": "default",
        "setting_name": "",
        "score": 43.3,
        "unit": "%",
        "printed": "43.3%",
        "reported_by": "google",
        "run_by": null,
        "label": "Gemini 3 Pro",
        "source": "https://blog.google/innovation-and-ai/models-and-research/gemini-models/gemini-3-1-pro/",
        "source_title": "Gemini 3.1 Pro: A smarter model for your most complex tasks",
        "source_date": "2026-02-19",
        "where": "benchmark table (image), row 'SWE-Bench Pro (Public)', column 'Gemini 3 Pro Thinking (High)'",
        "own": false,
        "primary": true,
        "parallel": false,
        "notes": "single attempt; Thinking (High)",
        "caveat": "",
        "human_verified": true,
        "human_verified_on": "2026-10-17",
        "check": "human",
        "corroborated": false,
        "corroborated_on": null,
        "corroborated_by": [],
        "source_image": "https://frontierlog.ai/sources/6b612deebae92383.gif",
        "source_image_original": "https://storage.googleapis.com/gweb-uniblog-publish-prod/original_images/gemini_3-1-pro__benchmarks.gif"
      }
    ]
  }
}