| Benchmark | Kimi K2 (Reasoning) | Llama 3.1 Instruct 405B |
|---|---|---|
| AA Agentic Index | 47.9 | 6.3 |
| AA Intelligence | 33.5 | 8.5 |
| AA-LCR | 70.3 | 24.7 |
| AA-Omniscience | -21.4 | -17.1 |
| aider_polyglot | 59.1 | 5.8 |
| Artificial Analysis Coding Index | 34.8 | 14.5 |
| critpt | 2.6 | 0 |
| gdpval | 24.5 | 0 |
| GPQA Diamond | 84.5 | 51.5 |
| HLE | 23.9 | 4.2 |
| IFBench | 68.1 | 39 |
| LiveCodeBench | 79.2 | 30.1 |
| longbench_v2 | 45.1 | 36.1 |
| mmlu_redux | 94.4 | 86.2 |
| MMLU-Pro | 84.6 | 73.4 |
| OmniScience Accuracy | 30.9 | 23.2 |
| OmniScience Non-Hallucination | 25.8 | 49 |
| scicode | 44.8 | 29.9 |
| simplebench | 39.6 | 23 |
| SWE-bench Verified | 71.3 | 24.5 |
| Terminal-Bench Hard | 31.1 | 6.8 |
| τ²-Bench Telecom (AA run) | 93 | 19 |
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.