Gemini 3.1 Pro
Preview model, and the last Gemini called Pro before Gemini 4. METR time horizon: about 6.4 hours.
- Announced
- Access
- Public
Benchmark scores
| Test | Score | Best on the tracker | How it was measured |
|---|---|---|---|
| Epoch Capabilities Indexmany tests combined into one score | 156 points | GPT-6.2 Astra 174 points | Independentlisted as "Gemini 3.1 Pro", CI 154.51 to 158.14Run 9 Nov 2026 |
| MirrorCoderebuilding whole programs | 8.9% | Claude Opus 5.5 79% | Independenthigh reasoning effort, MirrorCode (ML, +Private, 2L): 15 programs x 2 languages, 3 runs each, 10B-token / 7-day budget, ±4.6 SERun 10 Aug 2026 |
| APEX-Agentsprofessional office tasks | 35% | Gemini 4 Carbon 84% | Independentpass@1, default effort; listed as "Gemini 3.1 Pro" |
| Humanity's Last Examexpert questions, no tools | 46% | GPT-6 Astra 55% | Independentleaderboard entry "gemini-3.1-pro-preview (thinking high)" (gemini-3.1-pro-preview) |
| ARC-AGI-3puzzle games, standard harnessTypical first-time human player: 100% | 0.4% | Claude Fable 5.2 81% | IndependentARC Prize config "Gemini 3.1 Pro (Preview)" (google-gemini-3-1-pro-preview), standard harness |
| FrontierMath T1–3hard maths problems | 60% | Claude Fable 5.2 95% | Independentdefault settings (gemini-3.1-pro-preview)Run 11 Jun 2026 |
| GPQA DiamondPhD-level science questionsPhD experts: 70%Saturated | 94% | Gemini 4 Argon 96% | Independenthigh reasoning effort (gemini-3.1-pro-preview_high)Run 6 Aug 2026 |
| FrontierMath T4research-level maths problemsSaturated | 27% | GPT-6.1 Sol 100% | Independentdefault settings (gemini-3.1-pro-preview)Run 11 Jun 2026 |
| ARC-AGI-2visual puzzles, model aloneAverage person: 60%Saturated | 77% | GPT-6 Astra 95% | IndependentARC Prize config "Gemini 3.1 Pro (Preview)" (gemini-3-1-pro-preview) |
| OTIS Mock AIMEcompetition mathsSaturated | 96% | GPT-5.5 100% | Independentdefault settings (gemini-3.1-pro-preview)Run 20 Feb 2026 |
| SWE-bench Verifiedfixing real software bugsKnown flaws | 76% | Claude Opus 4.7 83% | Independentdefault settings, custom-tools endpoint (gemini-3.1-pro-preview-customtools)Run 24 Feb 2026 |
| METR time horizontask length an AI agent finishes half the timeNo longer updated | 6.4 h | Claude Mythos Preview 17 h | IndependentMETR Horizon v1.1 run 'gemini_3_1_pro'; 95% CI 233.5–694.8 min |
Also on these charts
Sources
- Gemini 3.1 Pro: Announcing our latest Gemini AI modelGoogle · Official · blog.google
- Gemini API release notesGoogle · Official · ai.google.dev
- METR time horizonsMETR · Official · metr.org
- Epoch AI Benchmarking Hub (model release dates and runs)Epoch AI · Dataset · 9 Nov 2026 · epoch.ai