| Benchmark | Opus 5 | Qwen3.6 27B |
|---|---|---|
| AA Agentic Index | 59.2 | 27.5 |
| AA Intelligence | 63.1 | 37.7 |
| AA-LCR | 78.7 | 73.3 |
| AA-Omniscience | 37.1 | -20 |
| Artificial Analysis Coding Index | 78 | 53.7 |
| critpt | 29.1 | 1.1 |
| DeepSWE 1.1 | 68.8 | 13.3 |
| gdpval | 67.2 | 32 |
| GPQA Diamond | 93.7 | 87.8 |
| HLE | 64.7 | 24 |
| livebench_agentic_coding | 65.2 | 39.3 |
| livebench_coding | 81.5 | 71.8 |
| livebench_data_analysis | 74.5 | 70.4 |
| livebench_instruction_following | 63.8 | 53.2 |
| livebench_language | 88.7 | 63.3 |
| livebench_math | 95.7 | 79.9 |
| livebench_reasoning | 91.2 | 70.3 |
| MMMU-Pro | 84.7 | 75.8 |
| OmniScience Accuracy | 60.9 | 19.6 |
| OmniScience Non-Hallucination | 40.5 | 50.7 |
| OSWorld-Verified | 4 | 63.9 |
| scicode | 55.7 | 39.8 |
| SWE-bench Multilingual | 89.5 | 71.3 |
| SWE-bench Pro | 79.2 | 53.5 |
| SWE-bench Verified | 96 | 77.2 |
| TauBench V3 - Banking | 44.7 | 16.7 |
| Terminal-Bench 2.1 | 89.1 | 60.7 |
Best tracked score per model per benchmark (default configuration; source-attributed). ↓ marks lower-is-better metrics. Open either model for its full surface, provenance and pricing.