| Benchmark | Claude Sonnet 4.6 | Qwen3.5 27B |
|---|---|---|
| AA Agentic Index | 61.6 | 54.6 |
| AA Intelligence | 48.4 | 34.6 |
| AA-LCR | 74 | 72.3 |
| AA-Omniscience | 12.2 | -44 |
| arena_vision | 1278 | 1221 |
| Artificial Analysis Coding Index | 63 | 34.9 |
| browsecomp | 76.2 | 61 |
| charxiv_rq | 72.4 | 79.5 |
| coding_arena_elo | 1523 | 1357 |
| critpt | 3.1 | 0.9 |
| gdpval | 54.8 | 33 |
| GPQA Diamond | 89.9 | 85.8 |
| HLE | 49 | 48.5 |
| IFBench | 56.6 | 76.5 |
| itbenchSre | 39.8 | 35.5 |
| MCP Atlas | 69.5 | 68.4 |
| mmmlu | 89.3 | 85.9 |
| MMMU-Pro | 75.6 | 75 |
| OmniScience Accuracy | 40.9 | 20.7 |
| OmniScience Non-Hallucination | 51.6 | 24.9 |
| OSWorld-Verified | 78.5 | 56.2 |
| scicode | 47 | 39.5 |
| SWE-bench Pro | 58.1 | 51.2 |
| SWE-bench Verified | 79.6 | 75 |
| Terminal-Bench 2.0 | 59.1 | 41.6 |
| Terminal-Bench Hard | 59.1 | 32.6 |
| vectara_answer_rate | 99.9 | 99.8 |
| vectara_avg_summary_length | 114.7 | 94.4 |
| vectara_factual_consistency | 89.4 | 87.9 |
| vectara_hallucination_rate ↓ | 10.6 | 12.1 |
| τ²-Bench Telecom (AA run) | 97.9 | 93.9 |
| τ³-Bench | 30.5 | 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.