| Benchmark | Gemma 4 E2B | Qwen3.6 35B A3B |
|---|---|---|
| AA Agentic Index | 7.4 | 21.6 |
| AA Intelligence | 10 | 32.1 |
| AA-LCR | 17 | 66.7 |
| AA-Omniscience | -23.6 | -22.2 |
| aime_2026 | 37.5 | 92.7 |
| Artificial Analysis Coding Index | 8.3 | 41.9 |
| Claw-Eval Avg | 53.1 | 68.7 |
| critpt | 0 | 0.3 |
| gdpval | 0 | 27.8 |
| GPQA Diamond | 43.4 | 86 |
| HallusionBench | 0.6 | 69.8 |
| HLE | 4.8 | 22.2 |
| IFBench | 38 | 64.4 |
| LiveCodeBench v6 | 44 | 80.4 |
| MathVista | 62.1 | 86.4 |
| MMLU-Pro | 60 | 85.6 |
| MMMU-Pro | 44.6 | 75.3 |
| ocrbench_v2 | 44.7 | 65.5 |
| OmniScience Accuracy | 8.2 | 18.8 |
| OmniScience Non-Hallucination | 67.6 | 49.5 |
| PinchBench | 44.2 | 88.1 |
| RealWorldQA | 56.2 | 85.3 |
| scicode | 20.9 | 35.8 |
| TauBench V3 - Banking | 4.5 | 9.3 |
| Terminal-Bench 2.1 | 0.4 | 44.9 |
| Terminal-Bench Hard | 3 | 34.8 |
| τ²-Bench Telecom (AA run) | 22.2 | 95.3 |
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.