| Benchmark | Kimi K2 (Reasoning) | Qwen3.6 35B A3B |
|---|---|---|
| AA Agentic Index | 47.9 | 21.6 |
| AA Intelligence | 33.5 | 32.1 |
| AA-LCR | 70.3 | 66.7 |
| AA-Omniscience | -21.4 | -22.2 |
| Artificial Analysis Coding Index | 34.8 | 41.9 |
| browsecomp | 60.2 | 48.7 |
| critpt | 2.6 | 0.3 |
| gdpval | 24.5 | 27.8 |
| GPQA (unspecified) | 74.2 | 86 |
| GPQA Diamond | 84.5 | 86 |
| HLE | 23.9 | 22.2 |
| HMMT 2025 | 89.4 | 89.1 |
| hmmt_feb_2025 | 93.3 | 90.7 |
| hmmt_nov_2025 | 89.2 | 89.1 |
| IFBench | 68.1 | 64.4 |
| imo_answer_bench | 78.6 | 78.9 |
| LiveCodeBench v6 | 83.1 | 80.4 |
| mmlu_redux | 94.4 | 93.3 |
| MMLU-Pro | 84.6 | 85.6 |
| OmniScience Accuracy | 30.9 | 18.8 |
| OmniScience Non-Hallucination | 25.8 | 49.5 |
| scicode | 44.8 | 35.8 |
| SWE-bench Multilingual | 61.1 | 67.2 |
| SWE-bench Verified | 71.3 | 73.4 |
| Terminal-Bench 2.0 | 35.7 | 51.5 |
| Terminal-Bench Hard | 31.1 | 34.8 |
| τ²-Bench Telecom (AA run) | 93 | 95.3 |
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.