| Benchmark | Kimi K2 (Reasoning) | Qwen3 VL 4B (Reasoning) |
|---|---|---|
| AA Agentic Index | 47.9 | 14.4 |
| AA Intelligence | 33.5 | 7.7 |
| AA-LCR | 70.3 | 23 |
| AA-Omniscience | -21.4 | -68.4 |
| AIME 2025 | 94.5 | 74.5 |
| Artificial Analysis Coding Index | 34.8 | 6.7 |
| critpt | 2.6 | 0 |
| gdpval | 24.5 | 13.8 |
| GPQA Diamond | 84.5 | 64.1 |
| HLE | 23.9 | 4.6 |
| HMMT 2025 | 89.4 | 53.1 |
| IFBench | 68.1 | 36.6 |
| LiveCodeBench | 79.2 | 51.3 |
| LiveCodeBench v6 | 83.1 | 51.3 |
| mmlu_redux | 94.4 | 86 |
| MMLU-Pro | 84.6 | 73.6 |
| OmniScience Accuracy | 30.9 | 12.2 |
| OmniScience Non-Hallucination | 25.8 | 8.2 |
| scicode | 44.8 | 17.1 |
| Terminal-Bench Hard | 31.1 | 1.5 |
| τ²-Bench Telecom (AA run) | 93 | 15.5 |
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.