Claude Opus 4.6 vs Gemini 3.1 Pro
Claude Opus 4.6 
Gemini 3.1 Pro Verdict
- Claude Opus 4.6 and Gemini 3.1 Pro are level on quality (63.1 vs 63.9).
- Claude Opus 4.6 is stronger in agents & tools, coding, composite, human preference, knowledge, maths.
- Gemini 3.1 Pro is stronger in instruction following, long context, multimodal, reasoning.
- Gemini 3.1 Pro is 2.2× cheaper ($4.50 vs $10.00 per 1M blended).
- Gemini 3.1 Pro streams 2.9× faster (110 vs 38 tokens per second).
| Metric | Claude Opus 4.6 | Gemini 3.1 Pro |
|---|---|---|
| BenchLeader Index | 63.1 | 63.9 |
| Agents & tools score | 68.8 | 63.7 |
| Coding score | 62.3 | 59.8 |
| Composite score | 69.2 | 67.2 |
| Human preference score | 64.6 | 59.6 |
| Instruction following score | 53.7 | 69.5 |
| Knowledge score | 70.5 | 66.1 |
| Long context score | 65.5 | 67.6 |
| Maths score | 65.6 | 60.7 |
| Multimodal score | 63.6 | 66.1 |
| Reasoning score | 62.3 | 69.5 |
| Blended price $/M | $10.00 | $4.50 |
| Output speed | 38 tok/s | 110 tok/s |
| Time to first answer | 2.0 s | 30.5 s |
| Context window | 1M | 1.0M |
| GPQA Diamond | 90.5% | 94.1% |
| FrontierMath Tiers 1–3 | – | 59.6% |
| FrontierMath Tier 4 | – | 26.8% |
| OTIS Mock AIME | 94.4% | 95.6% |
| SWE-bench Verified (Epoch) | 78.7% | – |
| Humanity's Last Exam | 19.0% | 46.4% |
| Terminal-Bench | 79.8% | 80.2% |
| SimpleBench | 67.6% | 79.6% |
| SciCode | – | 58.9% |
| Cybench | 93.0% | – |
| Remote Labor Index | 4.2% | – |
| WeirdML | 77.9% | 72.1% |
| APEX-Agents | 32.4% | 33.5% |
| FrontierCode | 26.6% | – |
| ProofBench | – | 26.0% |
| GSO-Bench | 33.3% | 22.6% |
| Epoch Capabilities Index | 155.3 | 155 |
| LMArena Text | 1498 | 1487 |
| LMArena Hard Prompts | 1527 | 1507 |
| LMArena Coding | 1546 | 1521 |
| LMArena WebDev | 1537 | 1447 |
| LMArena Vision | 1311 | 1295 |
| LMArena Agent | – | -5.3 |
| AA Intelligence Index | 31.9 | 30.4 |
| IFBench | 53.1% | 77.1% |
| AA-LCR | 78.0% | 82.0% |
| MMMU-Pro | 75.4% | 82.4% |
| AA-Omniscience | 13.7 | 31.9 |
| Terminal-Bench Hard | 48.5% | 53.8% |
| GPQA Diamond (AA) | 89.6% | 94.1% |
| Humanity's Last Exam (AA) | 39.9% | 47.0% |
| SciCode (AA) | – | 58.7% |
| τ²-Bench Telecom (AA) | 92.1% | 95.6% |
| 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 | 38.3% | 35.3% |
| TutorBench | – | 53.0% |
| EQ-Bench 4 | 1223 | 1142 |
| Kagi LLM Benchmark | 72.4% | – |
| SWE-bench Verified (bash only) | 75.6% | – |
| SWE-bench Verified (any scaffold) | 75.6% | – |
| ARC-AGI-1 | – | 98.0% |
| ARC-AGI-2 | – | 77.1% |
| ARC-AGI-3 | – | 0.4% |
Data as of 2026-09-09. Best configuration of each model; every score links to its source on the model pages.