Across 31 shared benchmarks, Claude Opus 4.8 scores higher on 30 and Qwen3.5 122B A10B on 1. The widest gap is OmniScience Non-Hallucination, where Claude Opus 4.8 scores 60.7 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 4.8 | Qwen3.5 122B A10B |
|---|---|---|
| AA Agentic Index | 49.4 | 21.3 |
| AA Intelligence | 57.3 | 32.8 |
| AA-LCR | 73 | 70.3 |
| AA-Omniscience | 28.8 | -41.5 |
| arena_vision | 1287 | 1246 |
| Artificial Analysis Coding Index | 74.3 | 45.7 |
| browsecomp | 88.5 | 63.8 |
| coding_arena_elo | 1539 | 1358 |
| critpt | 20.9 | 0.9 |
| gdpval | 54.1 | 24.3 |
| GPQA Diamond | 93.6 | 86.6 |
| HLE | 57.9 | 47.5 |
| IFBench | 62.2 | 76.1 |
| include | 87.6 | 82.8 |
| longbench_v2 | 69.1 | 60.2 |
| mathvision | 86.7 | 86.2 |
| MMMU-Pro | 78.9 | 76.9 |
| MMVU | 79.2 | 74.7 |
| OmniScience Accuracy | 48.8 | 24.4 |
| OmniScience Non-Hallucination | 60.7 | 12.9 |
| OSWorld-Verified | 83.4 | 58 |
| scicode | 53.5 | 42 |
| screenspot_pro_no_tools | 87.9 | 70.4 |
| SWE-bench Verified | 88.6 | 72 |
| TauBench V3 - Banking | 34.2 | 15.3 |
| Terminal-Bench 2.0 | 74.6 | 49.4 |
| Terminal-Bench 2.1 | 85 | 47.6 |
| Terminal-Bench Hard | 58.3 | 31.1 |
| WideSearch | 72.9 | 60.5 |
| τ²-Bench Telecom (AA run) | 94.4 | 93.6 |
| τ³-Bench | 27.6 | 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.