| Benchmark | Claude Sonnet 5 | Qwen3.5 27B |
|---|---|---|
| AA Agentic Index | 49.7 | 54.6 |
| AA Intelligence | 55.3 | 34.6 |
| AA-LCR | 77 | 72.3 |
| AA-Omniscience | 16.4 | -44 |
| Artificial Analysis Coding Index | 71.5 | 34.9 |
| browsecomp | 84.7 | 61 |
| charxiv_rq | 77 | 79.5 |
| coding_arena_elo | 1540 | 1357 |
| critpt | 16.9 | 0.9 |
| gdpval | 54.8 | 33 |
| GPQA Diamond | 91.1 | 85.8 |
| HLE | 57.4 | 48.5 |
| LVBench | 68.5 | 73.6 |
| MMMU-Pro | 77.3 | 75 |
| OmniScience Accuracy | 40 | 20.7 |
| OmniScience Non-Hallucination | 60.6 | 24.9 |
| OSWorld-Verified | 81.2 | 56.2 |
| scicode | 53.6 | 39.5 |
| SWE-bench Multilingual | 78.3 | 69.3 |
| SWE-bench Pro | 63.2 | 51.2 |
| SWE-bench Verified | 85.2 | 75 |
| Terminal-Bench 2.0 | 80.4 | 41.6 |
| τ³-Bench | 28.2 | 68.4 |
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.