| Benchmark | Claude Sonnet 5 | Qwen3.5 397B A17B |
|---|---|---|
| AA Agentic Index | 49.7 | 53.3 |
| AA Intelligence | 55.3 | 34.3 |
| AA-LCR | 77 | 72.7 |
| AA-Omniscience | 16.4 | -30.8 |
| arena_elo | 1463 | 1442 |
| arena_text_factuality | 1453 | 1445 |
| Artificial Analysis Coding Index | 71.5 | 48.2 |
| browsecomp | 84.7 | 69 |
| charxiv_rq | 77 | 80.8 |
| coding_arena_elo | 1540 | 1399 |
| critpt | 16.9 | 1.7 |
| gdpval | 54.8 | 35.8 |
| GDPval-AA v2 | 1618 | 962 |
| GPQA Diamond | 91.1 | 89.3 |
| HLE | 57.4 | 29 |
| MMMU-Pro | 77.3 | 79 |
| OmniScience Accuracy | 40 | 30.8 |
| OmniScience Non-Hallucination | 60.6 | 17.3 |
| scicode | 53.6 | 42 |
| simpleqa_verified | 25 | 26 |
| SWE-bench Multilingual | 78.3 | 69.3 |
| SWE-bench Pro | 63.2 | 50.9 |
| SWE-bench Verified | 85.2 | 76.4 |
| TauBench V3 - Banking | 37.3 | 13.4 |
| Terminal-Bench 2.0 | 80.4 | 52.5 |
| Terminal-Bench 2.1 | 80.5 | 51.3 |
| toolathlon | 54.3 | 38.3 |
| usamo_2026 | 79.5 | 36.3 |
| τ³-Bench | 28.2 | 13.4 |
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.