| Benchmark | Claude Opus 4.7 | Qwen3.5 27B |
|---|---|---|
| AA Agentic Index | 64.6 | 54.6 |
| AA Intelligence | 55 | 34.6 |
| AA-LCR | 75.3 | 72.3 |
| AA-Omniscience | 27.3 | -44 |
| aime_2026 | 95.8 | 92.6 |
| arena_vision | 1304 | 1221 |
| Artificial Analysis Coding Index | 73.6 | 34.9 |
| browsecomp | 79.8 | 61 |
| charxiv_rq | 82.1 | 79.5 |
| coding_arena_elo | 1557 | 1357 |
| critpt | 12 | 0.9 |
| gdpval | 58.6 | 33 |
| GPQA Diamond | 94.2 | 85.8 |
| HLE | 54.7 | 48.5 |
| hmmt_feb_2026 | 93.9 | 84.3 |
| IFBench | 58.6 | 76.5 |
| itbenchSre | 46.7 | 35.5 |
| MCP Atlas | 79.1 | 68.4 |
| mmmlu | 91.5 | 85.9 |
| MMMU-Pro | 78.8 | 75 |
| OmniScience Accuracy | 48.9 | 20.7 |
| OmniScience Non-Hallucination | 57.7 | 24.9 |
| OSWorld-Verified | 82.8 | 56.2 |
| scicode | 54.5 | 39.5 |
| screenspot_pro_no_tools | 79.5 | 70.3 |
| SWE-bench Multilingual | 80.5 | 69.3 |
| SWE-bench Pro | 64.3 | 51.2 |
| SWE-bench Verified | 87.6 | 75 |
| Terminal-Bench 2.0 | 69.4 | 41.6 |
| Terminal-Bench Hard | 54.5 | 32.6 |
| vectara_answer_rate | 98 | 99.8 |
| vectara_avg_summary_length | 149.1 | 94.4 |
| vectara_factual_consistency | 88 | 87.9 |
| vectara_hallucination_rate ↓ | 12 | 12.1 |
| τ²-Bench Telecom (AA run) | 88.6 | 93.9 |
| τ³-Bench | 28.9 | 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.