Skip to content

Google

Gemini 3 Pro

First Gemini 3 model, released as a preview. METR time horizon: about 3.7 hours.

Announced
Access
Public

Benchmark scores

TestScoreBest on the trackerHow it was measured
Epoch Capabilities Indexmany tests combined into one score153 pointsGPT-6.2 Astra 174 pointsIndependentlisted as "Gemini 3 Pro", CI 151.34 to 154.57Run 9 Nov 2026
Remote Labor Indexpaid freelance projects1.3%Claude Fable 5.2 41%Independentgemini-3-pro-previewRun 9 Dec 2025
Humanity's Last Examexpert questions, no tools38%GPT-6 Astra 55%Independentleaderboard entry "gemini-3-pro-preview" (gemini-3-pro-preview)
GPQA DiamondPhD-level science questionsPhD experts: 70%Saturated93%Gemini 4 Argon 96%Independentdefault settings (gemini-3-pro-preview)Run 19 Nov 2025
ARC-AGI-2visual puzzles, model aloneAverage person: 60%Saturated31%GPT-6 Astra 95%IndependentARC Prize config "Gemini 3 Pro" (gemini-3-pro-preview)
OTIS Mock AIMEcompetition mathsSaturated91%GPT-5.5 100%Independentdefault settings (gemini-3-pro-preview)Run 19 Nov 2025
SWE-bench Verifiedfixing real software bugsKnown flaws73%Claude Opus 4.7 83%Independentdefault settings (gemini-3-pro-preview)Run 13 Feb 2026
METR time horizontask length an AI agent finishes half the timeNo longer updated3.7 hClaude Mythos Preview 17 hIndependentMETR Horizon v1.1 run 'gemini_3_pro'; 95% CI 139.6–379.2 min

Compare every model on the same tests →

Also on these charts

Sources