Across 31 shared benchmarks, Claude Opus 4.7 scores higher on 26 and Qwen3.5 122B A10B on 5. The widest gap is OmniScience Non-Hallucination, where Claude Opus 4.7 scores 57.7 against 12.9. Qwen3.5 122B A10B is the cheaper of the two on tracked API pricing ($0.40 against $5.00 per million input tokens).
| Benchmark | Claude Opus 4.7 | Qwen3.5 122B A10B |
|---|---|---|
| AA Agentic Index | 64.6 | 21.3 |
| AA Intelligence | 55 | 32.8 |
| AA-LCR | 75.3 | 70.3 |
| AA-Omniscience | 27.3 | -41.5 |
| arena_vision | 1316 | 1246 |
| Artificial Analysis Coding Index | 73.6 | 45.7 |
| browsecomp | 79.8 | 63.8 |
| coding_arena_elo | 1557 | 1358 |
| critpt | 12 | 0.9 |
| gdpval | 58.6 | 24.3 |
| GPQA Diamond | 94.2 | 86.6 |
| HLE | 54.7 | 47.5 |
| IFBench | 58.6 | 76.1 |
| mmmlu | 91.5 | 86.7 |
| MMMU-Pro | 78.8 | 76.9 |
| OmniScience Accuracy | 48.9 | 24.4 |
| OmniScience Non-Hallucination | 57.7 | 12.9 |
| OSWorld-Verified | 82.8 | 58 |
| scicode | 54.5 | 42 |
| screenspot_pro_no_tools | 79.5 | 70.4 |
| SWE-bench Verified | 87.6 | 72 |
| TauBench V3 - Banking | 34.6 | 15.3 |
| Terminal-Bench 2.0 | 69.4 | 49.4 |
| Terminal-Bench 2.1 | 83.1 | 47.6 |
| Terminal-Bench Hard | 54.5 | 31.1 |
| vectara_answer_rate | 98 | 99.8 |
| vectara_avg_summary_length | 149.1 | 86.4 |
| vectara_factual_consistency | 88 | 88.8 |
| vectara_hallucination_rate ↓ | 12 | 11.2 |
| τ²-Bench Telecom (AA run) | 88.6 | 93.6 |
| τ³-Bench | 28.9 | 13.6 |
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. Quoted rates are the price-setter row we currently track for each model — its direct or vendor-official listing where one exists (anthropic-official, direct), otherwise the lowest tracked offer.