| Benchmark | Kimi K2 (Reasoning) | Qwen3.5 397B A17B |
|---|---|---|
| AA Agentic Index | 47.9 | 53.3 |
| AA Intelligence | 33.5 | 34.3 |
| AA-LCR | 70.3 | 72.7 |
| AA-Omniscience | -21.4 | -30.8 |
| Artificial Analysis Coding Index | 34.8 | 48.2 |
| browsecomp | 60.2 | 69 |
| browsecomp_zh | 62.3 | 70.3 |
| critpt | 2.6 | 1.7 |
| gdpval | 24.5 | 35.8 |
| GPQA Diamond | 84.5 | 89.3 |
| HLE | 23.9 | 29 |
| HLE (with tools) | 44.9 | 48.3 |
| HMMT 2025 | 89.4 | 92.7 |
| hmmt_feb_2025 | 93.3 | 94.8 |
| hmmt_nov_2025 | 89.2 | 92.7 |
| IFBench | 68.1 | 78.8 |
| imo_answer_bench | 78.6 | 80.9 |
| LiveCodeBench v6 | 83.1 | 83.6 |
| longbench_v2 | 45.1 | 63.2 |
| mmlu_redux | 94.4 | 94.9 |
| MMLU-Pro | 84.6 | 87.8 |
| multichallenge | 66.4 | 67.6 |
| OmniScience Accuracy | 30.9 | 30.8 |
| OmniScience Non-Hallucination | 25.8 | 17.3 |
| scicode | 44.8 | 42 |
| Seal-0 | 56.3 | 46.9 |
| SWE-bench Multilingual | 61.1 | 69.3 |
| SWE-bench Verified | 71.3 | 76.4 |
| Terminal-Bench 2.0 | 35.7 | 52.5 |
| Terminal-Bench Hard | 31.1 | 40.9 |
| τ²-Bench Telecom (AA run) | 93 | 95.6 |
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.