Claude Opus 4.6 vs GPT-5.5
Claude Opus 4.6 
GPT-5.5 Verdict
- GPT-5.5 leads on quality: 66.6 vs 63.1.
- Claude Opus 4.6 is stronger in agents & tools, coding, maths.
- GPT-5.5 is stronger in composite, human preference, instruction following, knowledge, long context, multimodal, reasoning.
- They cost about the same ($10.00 per 1M blended).
- GPT-5.5 streams 2.3× faster (88 vs 38 tokens per second).
| Metric | Claude Opus 4.6 | GPT-5.5 |
|---|---|---|
| BenchLeader Index | 63.1 | 66.6 |
| Agents & tools score | 68.8 | 68.4 |
| Coding score | 62.3 | 57.3 |
| Composite score | 69.2 | 77.7 |
| Human preference score | 64.6 | 67.6 |
| Instruction following score | 53.7 | 73.3 |
| Knowledge score | 70.5 | 73.7 |
| Long context score | 65.5 | 68.8 |
| Maths score | 65.6 | – |
| Multimodal score | 63.6 | 64.9 |
| Reasoning score | 62.3 | 69.7 |
| Blended price $/M | $10.00 | $11.25 |
| Output speed | 38 tok/s | 88 tok/s |
| Time to first answer | 2.0 s | 62.3 s |
| Context window | 1M | 1.1M |
| GPQA Diamond | 90.5% | – |
| OTIS Mock AIME | 94.4% | – |
| SWE-bench Verified (Epoch) | 78.7% | – |
| Humanity's Last Exam | 19.0% | – |
| Terminal-Bench | 79.8% | 84.7% |
| SimpleBench | 67.6% | 69.0% |
| Cybench | 93.0% | – |
| Remote Labor Index | 4.2% | 6.3% |
| WeirdML | 77.9% | – |
| APEX-Agents | 32.4% | 38.5% |
| FrontierCode | 26.6% | 43.0% |
| GSO-Bench | 33.3% | – |
| Epoch Capabilities Index | 155.3 | 159.1 |
| LMArena Text | 1498 | 1477 |
| LMArena Hard Prompts | 1527 | 1498 |
| LMArena Coding | 1546 | 1509 |
| LMArena WebDev | 1537 | 1458 |
| LMArena Vision | 1311 | 1296 |
| LMArena Agent | – | 3 |
| AA Intelligence Index | 31.9 | 38.6 |
| IFBench | 53.1% | 75.8% |
| AA-LCR | 78.0% | 84.3% |
| MMMU-Pro | 75.4% | 79.9% |
| AA-Omniscience | 13.7 | 20.5 |
| Terminal-Bench Hard | 48.5% | 60.6% |
| GPQA Diamond (AA) | 89.6% | 93.5% |
| Humanity's Last Exam (AA) | 39.9% | 45.8% |
| SciCode (AA) | – | 55.8% |
| τ²-Bench Telecom (AA) | 92.1% | 93.9% |
| HiL-Bench | 38.3% | 39.7% |
| EQ-Bench 4 | 1223 | 1315 |
| Kagi LLM Benchmark | 72.4% | 88.8% |
| SWE-bench Verified (bash only) | 75.6% | – |
| SWE-bench Verified (any scaffold) | 75.6% | – |
Data as of 2026-09-09. Best configuration of each model; every score links to its source on the model pages.