| Benchmark | Gemma 4 26B A4B | Kimi K2 (Reasoning) |
|---|---|---|
| AA Agentic Index | 32.1 | 47.9 |
| AA Intelligence | 26.1 | 33.5 |
| AA-LCR | 61.7 | 70.3 |
| AA-Omniscience | -50.8 | -21.4 |
| AIME 2025 | 90 | 94.5 |
| Artificial Analysis Coding Index | 39.3 | 34.8 |
| browsecomp | 26.3 | 60.2 |
| critpt | 0 | 2.6 |
| gdpval | 25.7 | 24.5 |
| GPQA (unspecified) | 82.3 | 74.2 |
| GPQA Diamond | 82.3 | 84.5 |
| HLE | 19.3 | 23.9 |
| HLE (with tools) | 17.2 | 44.9 |
| hmmt_feb_2025 | 91.7 | 93.3 |
| hmmt_nov_2025 | 87.5 | 89.2 |
| IFBench | 72.4 | 68.1 |
| imo_answer_bench | 74.3 | 78.6 |
| LiveCodeBench | 79.8 | 79.2 |
| LiveCodeBench v6 | 77.1 | 83.1 |
| mmlu_redux | 92.7 | 94.4 |
| MMLU-Pro | 85.2 | 84.6 |
| OmniScience Accuracy | 19.1 | 30.9 |
| OmniScience Non-Hallucination | 13.6 | 25.8 |
| scicode | 40.3 | 44.8 |
| SWE-bench Multilingual | 43.4 | 61.1 |
| SWE-bench Verified | 57.4 | 71.3 |
| Terminal-Bench 2.0 | 34.2 | 35.7 |
| Terminal-Bench Hard | 25 | 31.1 |
| τ²-Bench | 68.2 | 74.3 |
| τ²-Bench Telecom (AA run) | 43.6 | 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.