| Benchmark | Gemma 4 E2B | Kimi K2 (Reasoning) |
|---|---|---|
| AA Agentic Index | 7.4 | 47.9 |
| AA Intelligence | 10 | 33.5 |
| AA-LCR | 17 | 70.3 |
| AA-Omniscience | -23.6 | -21.4 |
| AIME25 no tools | 26.3 | 94.5 |
| Artificial Analysis Coding Index | 8.3 | 34.8 |
| critpt | 0 | 2.6 |
| gdpval | 0 | 24.5 |
| GPQA Diamond | 43.4 | 84.5 |
| HLE | 4.8 | 23.9 |
| IFBench | 38 | 68.1 |
| LiveCodeBench v6 | 44 | 83.1 |
| LiveCodeBenchV6 no tools | 54.9 | 83.1 |
| MMLU-Pro | 60 | 84.6 |
| OmniScience Accuracy | 8.2 | 30.9 |
| OmniScience Non-Hallucination | 67.6 | 25.8 |
| scicode | 20.9 | 44.8 |
| Terminal-Bench Hard | 3 | 31.1 |
| τ²-Bench | 24.5 | 74.3 |
| τ²-Bench Telecom (AA run) | 22.2 | 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.