| Benchmark | DeepSeek-V3 | Kimi K2 (Reasoning) |
|---|---|---|
| AA Agentic Index | 1.6 | 47.9 |
| AA Intelligence | 15.4 | 33.5 |
| AA-LCR | 41.3 | 70.3 |
| AA-Omniscience | -37.6 | -21.4 |
| aider_polyglot | 55.1 | 59.1 |
| AIME 2025 | 51.3 | 94.5 |
| Artificial Analysis Coding Index | 23 | 34.8 |
| critpt | 0 | 2.6 |
| Frames | 73.3 | 87 |
| gdpval | 0 | 24.5 |
| GPQA Diamond | 68.4 | 84.5 |
| HLE | 5.2 | 23.9 |
| HMMT 2025 | 27.5 | 89.4 |
| hmmt_feb_2025 | 29.2 | 93.3 |
| IFBench | 41 | 68.1 |
| LiveCodeBench | 49.6 | 79.2 |
| LiveCodeBench v6 | 46.9 | 83.1 |
| longbench_v2 | 48.7 | 45.1 |
| mmlu_redux | 90.5 | 94.4 |
| MMLU-Pro | 81.2 | 84.6 |
| multichallenge | 31.4 | 66.4 |
| OmniScience Accuracy | 25.4 | 30.9 |
| OmniScience Non-Hallucination | 23.3 | 25.8 |
| scicode | 35.8 | 44.8 |
| simplebench | 27.2 | 39.6 |
| SWE-bench Verified | 42 | 71.3 |
| Terminal-Bench Hard | 15.2 | 31.1 |
| vectara_answer_rate | 97.5 | 98.6 |
| vectara_avg_summary_length | 81.7 | 59.2 |
| vectara_factual_consistency | 93.9 | 82.1 |
| vectara_hallucination_rate ↓ | 6.1 | 17.9 |
| τ²-Bench | 32.5 | 74.3 |
| τ²-Bench Telecom (AA run) | 47.1 | 93 |
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.