| Benchmark | Opus 5 | Qwen3.5 27B |
|---|---|---|
| AA Agentic Index | 59.2 | 54.6 |
| AA Intelligence | 63.1 | 34.6 |
| AA-LCR | 78.7 | 72.3 |
| AA-Omniscience | 37.1 | -44 |
| Artificial Analysis Coding Index | 78 | 34.9 |
| browsecomp | 90.8 | 61 |
| coding_arena_elo | 1662 | 1357 |
| critpt | 29.1 | 0.9 |
| gdpval | 67.2 | 33 |
| GPQA Diamond | 93.7 | 85.8 |
| HLE | 64.7 | 48.5 |
| MCP Atlas | 85.8 | 68.4 |
| MMMU-Pro | 84.7 | 75 |
| OmniScience Accuracy | 60.9 | 20.7 |
| OmniScience Non-Hallucination | 40.5 | 24.9 |
| OSWorld-Verified | 4 | 56.2 |
| scicode | 55.7 | 39.5 |
| SWE-bench Multilingual | 89.5 | 69.3 |
| SWE-bench Pro | 79.2 | 51.2 |
| SWE-bench Verified | 96 | 75 |
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.