Across 35 shared benchmarks, Kimi K2.6 scores higher on 32 and Qwen3.5 122B A10B on 3. The widest gap is critpt, where Kimi K2.6 scores 8 against 0.9. Qwen3.5 122B A10B is the cheaper of the two on tracked API pricing ($0.40 against $0.91 per million input tokens).
| Benchmark | Kimi K2.6 | Qwen3.5 122B A10B |
|---|---|---|
| AA Agentic Index | 58.7 | 21.3 |
| AA Intelligence | 45.1 | 32.8 |
| AA-LCR | 76.7 | 70.3 |
| AA-Omniscience | 6.4 | -41.5 |
| arena_vision | 1281 | 1246 |
| Artificial Analysis Coding Index | 61.8 | 45.7 |
| baby_vision | 68.5 | 40.2 |
| browsecomp | 86.3 | 63.8 |
| coding_arena_elo | 1509 | 1358 |
| critpt | 8 | 0.9 |
| gdpval | 41.4 | 24.3 |
| GPQA Diamond | 91.1 | 86.6 |
| HLE | 37.5 | 47.5 |
| IFBench | 76 | 76.1 |
| LiveCodeBench v6 | 89.6 | 78.9 |
| mathvision | 93.2 | 86.2 |
| MMLU-Pro | 87.1 | 86.7 |
| MMMU-Pro | 80.1 | 76.9 |
| OJBench | 60.6 | 39.5 |
| OmniScience Accuracy | 32.8 | 24.4 |
| OmniScience Non-Hallucination | 60.7 | 12.9 |
| OSWorld-Verified | 73.1 | 58 |
| scicode | 53.5 | 42 |
| SWE-bench Verified | 80.2 | 72 |
| TauBench V3 - Banking | 23.3 | 15.3 |
| Terminal-Bench 2.0 | 66.7 | 49.4 |
| Terminal-Bench 2.1 | 65.9 | 47.6 |
| Terminal-Bench Hard | 43.9 | 31.1 |
| vectara_answer_rate | 99.7 | 99.8 |
| vectara_avg_summary_length | 116.7 | 86.4 |
| vectara_factual_consistency | 89.2 | 88.8 |
| vectara_hallucination_rate ↓ | 10.8 | 11.2 |
| WideSearch | 80.8 | 60.5 |
| τ²-Bench Telecom (AA run) | 95.9 | 93.6 |
| τ³-Bench | 20.6 | 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 (moonshot-official, direct), otherwise the lowest tracked offer.