Up to four models
Gemini 3.1 Pro vs Gemini 3 Pro
On the index (GPT-4 = 100), Gemini 3.1 Pro scores 166 and Gemini 3 Pro 164. Side by side on the same third-party tests, with price and context. The link in your address bar shares this exact comparison.
Gemini 3.1 Pro166150–183
WeightsClosed
ReleasedFeb 2026
Price in / out$2.0 / $12
Context1.05M
Core tests taken9 of 9
Gemini 3.1 Pro and Gemini 3 Pro are too close to call (3 points apart, within the uncertainty), ahead on 5 of 5 shared core tests.
Test by testthird-party results
Gemini 3.1 ProGemini 3 ProEvery result as a table
| Test | Gemini 3.1 Pro | Gemini 3 Pro |
|---|---|---|
| Reasoning & knowledge | ||
| GPQA Diamond (Epoch AI) | 94.4% | 92.6% |
| Humanity's Last Exam | 46.4% | 37.5% |
| GPQA Diamond (Vals AI) | 95.5% | 91.7% |
| MMLU Pro | 91.0% | 90.1% |
| MMMU Pro | 88.2% | 87.5% |
| CL-bench | 20.8% | 15.8% |
| SimpleBench | 79.6% | 76.4% |
| Maths | ||
| AIME-style maths (OTIS mock) | 95.6% | 91.4% |
| FrontierMath (2025 set) | 36.9% | 37.6% |
| FrontierMath tier 4 (2025 set) | 16.7% | 18.8% |
| Coding | ||
| SWE-bench Verified | 75.6% | 72.9% |
| SWE-bench Pro (public set) | 46.1% | 43.3% |
| LiveCodeBench | 88.5% | 86.4% |
| SWE-bench | 78.8% | 76.4% |
| Vibe Code Bench v1.1 | 32.0% | 14.3% |
| GSO-Bench | 22.6% | 18.6% |
| WeirdML | 72.1% | 69.9% |
| Agents | ||
| BALROG | 57.0% | 58.1% |
| DeepResearch Bench | 47.8% | 46.3% |
| Professional work | ||
| CorpFin v2 | 64.5% | 63.7% |
| LegalBench | 87.4% | 87.0% |
| MedCode | 59.1% | 52.2% |
| MedScribe | 76.1% | 72.0% |
| MortgageTax | 69.4% | 69.1% |
| SAGE | 48.7% | 47.6% |
| TaxEval v2 | 72.9% | 72.6% |
| Novel problems | ||
| ARC-AGI-2 | 77.1% | 54.0% |
| ARC-AGI-1 | 98.0% | 75.0% |
Task length · METR46 sec43 sec
46 sec43 secTests at least two of these models have taken. Bold rows are the core tests.
