Gemini 3.1 Pro vs GPT-6 Astra
Gemini 3.1 Pro 
GPT-6 Astra Verdict
- GPT-6 Astra (max) leads on quality: 70.6 vs 63.9.
- Gemini 3.1 Pro is stronger in human preference, instruction following, long context, multimodal.
- GPT-6 Astra (max) is stronger in agents & tools, coding, composite, knowledge, maths, reasoning.
- Gemini 3.1 Pro is 4.9× cheaper ($4.50 vs $22.00 per 1M blended).
- Gemini 3.1 Pro streams 2.0× faster (110 vs 54 tokens per second).
| Metric | Gemini 3.1 Pro | GPT-6 Astra (max) |
|---|---|---|
| BenchLeader Index | 63.9 | 70.6 |
| Agents & tools score | 63.7 | 68.2 |
| Coding score | 59.8 | 79.0 |
| Composite score | 67.2 | 73.1 |
| Human preference score | 59.6 | – |
| Instruction following score | 69.5 | – |
| Knowledge score | 66.1 | 79.8 |
| Long context score | 67.6 | – |
| Maths score | 60.7 | 77.0 |
| Multimodal score | 66.1 | – |
| Reasoning score | 69.5 | 70.9 |
| Blended price $/M | $4.50 | $22.00 |
| Output speed | 110 tok/s | 54 tok/s |
| Time to first answer | 30.5 s | 328.6 s |
| Context window | 1.0M | 1.1M |
| GPQA Diamond | 94.1% | 95.8% |
| FrontierMath Tiers 1–3 | 59.6% | 93.7% |
| FrontierMath Tier 4 | 26.8% | 97.6% |
| OTIS Mock AIME | 95.6% | 100.0% |
| SimpleQA Verified | – | 75.6% |
| Humanity's Last Exam | 46.4% | – |
| Terminal-Bench | 80.2% | 58.2% |
| SimpleBench | 79.6% | – |
| SciCode | 58.9% | 56.5% |
| WeirdML | 72.1% | – |
| APEX-Agents | 33.5% | – |
| FrontierCode | – | 53.3% |
| ProofBench | 26.0% | – |
| GSO-Bench | 22.6% | – |
| Epoch Capabilities Index | 155 | – |
| LMArena Text | 1487 | – |
| LMArena Hard Prompts | 1507 | – |
| LMArena Coding | 1521 | – |
| LMArena WebDev | 1447 | 1796 |
| LMArena Vision | 1295 | – |
| LMArena Agent | -5.3 | 12.5 |
| LiveBench | – | 82.2% |
| LiveBench Reasoning | – | 92.7% |
| LiveBench Coding | – | 80.4% |
| LiveBench Agentic Coding | – | 57.3% |
| LiveBench Mathematics | – | 96.8% |
| LiveBench Data Analysis | – | 83.0% |
| LiveBench Language | – | 89.4% |
| AA Intelligence Index | 30.4 | – |
| IFBench | 77.1% | – |
| AA-LCR | 82.0% | – |
| MMMU-Pro | 82.4% | – |
| AA-Omniscience | 31.9 | – |
| Terminal-Bench Hard | 53.8% | – |
| GPQA Diamond (AA) | 94.1% | – |
| Humanity's Last Exam (AA) | 47.0% | – |
| SciCode (AA) | 58.7% | – |
| τ²-Bench Telecom (AA) | 95.6% | – |
| Terminal-Bench 2.1 (Vals) | – | 87.3% |
| Vals Index | – | 66.6 |
| AIME 2026 | 98.3% | – |
| HMMT February 2026 | 94.7% | – |
| MathArena Apex | 60.9% | – |
| MultiChallenge | 71.4% | – |
| PRBench Finance | 41.9% | – |
| PRBench Legal | 44.0% | – |
| MultiNRC | 64.7% | – |
| HiL-Bench | 35.3% | – |
| TutorBench | 53.0% | – |
| EQ-Bench 4 | 1142 | – |
| ARC-AGI-1 | 98.0% | 97.5% |
| ARC-AGI-2 | 77.1% | 95.0% |
| ARC-AGI-3 | 0.4% | 62.7% |
Data as of 2026-09-09. Best configuration of each model; every score links to its source on the model pages.