Across 45 shared benchmarks, Qwen3.5 122B A10B scores higher on 20 and Qwen3.6 27B on 24, with 1 level. The widest gap is OmniScience Non-Hallucination, where Qwen3.6 27B scores 50.7 against 12.9. Qwen3.5 122B A10B is the cheaper of the two on tracked API pricing ($0.40 against $0.60 per million input tokens).
| Benchmark | Qwen3.5 122B A10B | Qwen3.6 27B |
|---|---|---|
| AA Agentic Index | 21.3 | 27.5 |
| AA Intelligence | 32.8 | 37.7 |
| AA-LCR | 70.3 | 73.3 |
| AA-Omniscience | -41.5 | -20 |
| Artificial Analysis Coding Index | 45.7 | 53.7 |
| baby_vision | 40.2 | 28.9 |
| C-Eval | 91.9 | 91.4 |
| CC-OCR | 81.8 | 81.2 |
| critpt | 0.9 | 1.1 |
| DynaMath | 85.9 | 85.6 |
| EmbSpatial-Bench | 0.8 | 84.6 |
| ERQA | 62 | 62.5 |
| gdpval | 24.3 | 32 |
| GPQA Diamond | 86.6 | 87.8 |
| HLE | 47.5 | 24 |
| HMMT 2025 | 90.3 | 90.7 |
| IFBench | 76.1 | 69.1 |
| LiveCodeBench v6 | 78.9 | 83.9 |
| mathvision | 86.2 | 85.1 |
| MathVista | 87.4 | 87.4 |
| MLVU | 87.3 | 86.6 |
| mmlu_redux | 94 | 93.5 |
| MMLU-Pro | 86.7 | 86.2 |
| MMMU | 83.9 | 82.9 |
| MMMU-Pro | 76.9 | 75.8 |
| MMStar | 82.9 | 81.4 |
| MV-Bench | 76.6 | 75.5 |
| OCRBench | 92.1 | 89.4 |
| OmniDocBench 1.5 | 89.8 | 89.4 |
| OmniScience Accuracy | 24.4 | 19.6 |
| OmniScience Non-Hallucination | 12.9 | 50.7 |
| OSWorld-Verified | 58 | 63.9 |
| RealWorldQA | 85.1 | 84.1 |
| RefSpatial-Bench | 0.7 | 70 |
| scicode | 42 | 39.8 |
| SimpleVQA | 0.6 | 56.1 |
| supergpqa | 67.1 | 66 |
| SWE-bench Verified | 72 | 77.2 |
| TauBench V3 - Banking | 15.3 | 16.7 |
| Terminal-Bench 2.0 | 49.4 | 59.3 |
| Terminal-Bench 2.1 | 47.6 | 60.7 |
| Terminal-Bench Hard | 31.1 | 34.8 |
| VideoMMMU | 82 | 84.4 |
| τ²-Bench Telecom (AA run) | 93.6 | 94.2 |
| τ³-Bench | 13.6 | 15.3 |
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 (alibaba-official, alibaba-official), otherwise the lowest tracked offer.