Across 38 shared benchmarks, Claude Sonnet 4 scores higher on 11 and Qwen3.5 122B A10B on 26, with 1 level. The widest gap is HMMT 2025, where Qwen3.5 122B A10B scores 90.3 against 15.9. Qwen3.5 122B A10B is the cheaper of the two on tracked API pricing ($0.40 against $3.00 per million input tokens).
| Benchmark | Claude Sonnet 4 | Qwen3.5 122B A10B |
|---|---|---|
| AA Agentic Index | 39.2 | 21.3 |
| AA Intelligence | 29.8 | 32.8 |
| AA-LCR | 69.7 | 70.3 |
| AA-Omniscience | 0.2 | -41.5 |
| arena_vision | 1207 | 1246 |
| Artificial Analysis Coding Index | 37.6 | 45.7 |
| browsecomp | 12.2 | 63.8 |
| browsecomp_zh | 29.1 | 69.9 |
| critpt | 1.1 | 0.9 |
| gdpval | 31.2 | 24.3 |
| GPQA Diamond | 78 | 86.6 |
| HLE | 10.7 | 47.5 |
| HMMT 2025 | 15.9 | 90.3 |
| IFBench | 55 | 76.1 |
| ifeval | 87.6 | 93.4 |
| LiveCodeBench v6 | 48.5 | 78.9 |
| mmlu_redux | 93.6 | 94 |
| MMLU-Pro | 84 | 86.7 |
| mmmlu | 86.5 | 86.7 |
| MMMU | 74.4 | 83.9 |
| MMMU-Pro | 62.4 | 76.9 |
| multichallenge | 57.1 | 61.5 |
| OJBench | 15.3 | 39.5 |
| OmniScience Accuracy | 22.7 | 24.4 |
| OmniScience Non-Hallucination | 70.9 | 12.9 |
| OSWorld-Verified | 42.2 | 58 |
| scicode | 40 | 42 |
| supergpqa | 55.7 | 67.1 |
| SWE-bench Verified | 80.2 | 72 |
| TauBench V3 - Banking | 16.7 | 15.3 |
| Terminal-Bench 2.1 | 36.3 | 47.6 |
| Terminal-Bench Hard | 31.1 | 31.1 |
| vectara_answer_rate | 98.6 | 99.8 |
| vectara_avg_summary_length | 145.8 | 86.4 |
| vectara_factual_consistency | 89.7 | 88.8 |
| vectara_hallucination_rate ↓ | 10.3 | 11.2 |
| τ²-Bench Telecom (AA run) | 64.6 | 93.6 |
| τ³-Bench | 13.8 | 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.