Across 38 shared benchmarks, Claude Opus 4.6 scores higher on 26 and Qwen3.5 122B A10B on 11, with 1 level. The widest gap is τ³-Bench, where Claude Opus 4.6 scores 72.4 against 13.6. 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 4.6 | Qwen3.5 122B A10B |
|---|---|---|
| AA Agentic Index | 67.6 | 21.3 |
| AA Intelligence | 44.9 | 32.8 |
| AA-LCR | 74.3 | 70.3 |
| AA-Omniscience | 13.7 | -41.5 |
| arena_vision | 1311 | 1246 |
| Artificial Analysis Coding Index | 48.1 | 45.7 |
| baby_vision | 14.8 | 40.2 |
| browsecomp | 86.8 | 63.8 |
| coding_arena_elo | 1545 | 1358 |
| critpt | 12.6 | 0.9 |
| ERQA | 40.8 | 62 |
| gdpval | 55.9 | 24.3 |
| GPQA Diamond | 91.3 | 86.6 |
| GSM8K | 0.6 | 94.5 |
| HLE | 53.1 | 47.5 |
| IFBench | 62.5 | 76.1 |
| LiveCodeBench v6 | 88.8 | 78.9 |
| mathvision | 71.2 | 86.2 |
| MMLU-Pro | 89.1 | 86.7 |
| mmmlu | 91.1 | 86.7 |
| MMMU-Pro | 77.3 | 76.9 |
| multichallenge | 56 | 61.5 |
| OmniDocBench 1.5 | 86.6 | 89.8 |
| OmniScience Accuracy | 47 | 24.4 |
| OmniScience Non-Hallucination | 37.2 | 12.9 |
| OSWorld-Verified | 72.7 | 58 |
| RealWorldQA | 73.9 | 85.1 |
| scicode | 52 | 42 |
| screenspot_pro_no_tools | 57.7 | 70.4 |
| SWE-bench Verified | 80.8 | 72 |
| Terminal-Bench 2.0 | 65.4 | 49.4 |
| Terminal-Bench Hard | 65.4 | 31.1 |
| vectara_answer_rate | 99.8 | 99.8 |
| vectara_avg_summary_length | 137.6 | 86.4 |
| vectara_factual_consistency | 87.8 | 88.8 |
| vectara_hallucination_rate ↓ | 12.2 | 11.2 |
| τ²-Bench Telecom (AA run) | 99.3 | 93.6 |
| τ³-Bench | 72.4 | 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.