Across 30 shared benchmarks, Claude Sonnet 4.6 scores higher on 28 and Qwen3.5 122B A10B on 2. The widest gap is OmniScience Non-Hallucination, where Claude Sonnet 4.6 scores 51.6 against 12.9. 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 4.6 | Qwen3.5 122B A10B |
|---|---|---|
| AA Agentic Index | 61.6 | 21.3 |
| AA Intelligence | 48.4 | 32.8 |
| AA-LCR | 74 | 70.3 |
| AA-Omniscience | 12.2 | -41.5 |
| arena_vision | 1281 | 1246 |
| Artificial Analysis Coding Index | 63 | 45.7 |
| browsecomp | 76.2 | 63.8 |
| coding_arena_elo | 1522 | 1358 |
| critpt | 3.1 | 0.9 |
| gdpval | 54.8 | 24.3 |
| GPQA Diamond | 89.9 | 86.6 |
| HLE | 49 | 47.5 |
| IFBench | 56.6 | 76.1 |
| mmmlu | 89.3 | 86.7 |
| MMMU-Pro | 75.6 | 76.9 |
| OmniScience Accuracy | 40.9 | 24.4 |
| OmniScience Non-Hallucination | 51.6 | 12.9 |
| OSWorld-Verified | 78.5 | 58 |
| scicode | 47 | 42 |
| SWE-bench Verified | 79.6 | 72 |
| TauBench V3 - Banking | 34.4 | 15.3 |
| Terminal-Bench 2.0 | 59.1 | 49.4 |
| Terminal-Bench 2.1 | 71.2 | 47.6 |
| Terminal-Bench Hard | 59.1 | 31.1 |
| vectara_answer_rate | 99.9 | 99.8 |
| vectara_avg_summary_length | 114.7 | 86.4 |
| vectara_factual_consistency | 89.4 | 88.8 |
| vectara_hallucination_rate ↓ | 10.6 | 11.2 |
| τ²-Bench Telecom (AA run) | 97.9 | 93.6 |
| τ³-Bench | 30.5 | 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, alibaba-official), otherwise the lowest tracked offer.