| Benchmark | Gemma 4 E2B | Qwen3.5 397B A17B |
|---|---|---|
| AA Agentic Index | 7.4 | 53.3 |
| AA Intelligence | 10 | 34.3 |
| AA-LCR | 17 | 72.7 |
| AA-Omniscience | -23.6 | -30.8 |
| aime_2026 | 37.5 | 94.2 |
| Artificial Analysis Coding Index | 8.3 | 48.2 |
| bfcl_v4 | 37 | 72.9 |
| Claw-Eval Avg | 53.1 | 70.7 |
| critpt | 0 | 1.7 |
| gdpval | 0 | 35.8 |
| GPQA Diamond | 43.4 | 89.3 |
| HLE | 4.8 | 29 |
| IFBench | 38 | 78.8 |
| ifeval | 82.9 | 92.6 |
| LiveCodeBench v6 | 44 | 83.6 |
| MMLU-Pro | 60 | 87.8 |
| mmmlu | 67.4 | 88.5 |
| MMMU-Pro | 44.6 | 79 |
| MMStar | 57.9 | 83.8 |
| OmniScience Accuracy | 8.2 | 30.8 |
| OmniScience Non-Hallucination | 67.6 | 17.3 |
| RealWorldQA | 56.2 | 83.9 |
| scicode | 20.9 | 42 |
| TauBench V3 - Banking | 4.5 | 13.4 |
| Terminal-Bench 2.1 | 0.4 | 51.3 |
| Terminal-Bench Hard | 3 | 40.9 |
| τ²-Bench Telecom (AA run) | 22.2 | 95.6 |
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.