| Benchmark | DeepSeek-V4-Pro | Kimi K2 (Reasoning) |
|---|---|---|
| AA Agentic Index | 63.3 | 47.9 |
| AA Intelligence | 53 | 33.5 |
| AA-LCR | 70 | 70.3 |
| AA-Omniscience | -10.6 | -21.4 |
| Artificial Analysis Coding Index | 59.4 | 34.8 |
| browsecomp | 83.4 | 60.2 |
| critpt | 13 | 2.6 |
| gdpval | 49 | 24.5 |
| GPQA Diamond | 90.5 | 84.5 |
| HLE | 48.2 | 23.9 |
| HLE (with tools) | 48.2 | 44.9 |
| hmmt_nov_2025 | 94.4 | 89.2 |
| IFBench | 76.5 | 68.1 |
| imo_answer_bench | 89.8 | 78.6 |
| LiveCodeBench | 93.5 | 79.2 |
| MMLU-Pro | 87.5 | 84.6 |
| OmniScience Accuracy | 43 | 30.9 |
| OmniScience Non-Hallucination | 12.2 | 25.8 |
| scicode | 50 | 44.8 |
| simplebench | 50.9 | 39.6 |
| SWE-bench Multilingual | 76.2 | 61.1 |
| SWE-bench Verified | 80.6 | 71.3 |
| Terminal-Bench 2.0 | 67.9 | 35.7 |
| Terminal-Bench Hard | 46.2 | 31.1 |
| vectara_answer_rate | 97.2 | 98.6 |
| vectara_avg_summary_length | 153.8 | 59.2 |
| vectara_factual_consistency | 91.4 | 82.1 |
| vectara_hallucination_rate ↓ | 8.6 | 17.9 |
| τ²-Bench Telecom (AA run) | 96.2 | 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.