| Benchmark | Claude Fable 5 | Gemma 4 E2B |
|---|---|---|
| AA Agentic Index | 56.6 | 7.4 |
| AA Intelligence | 62.1 | 10 |
| AA-LCR | 76.7 | 17 |
| AA-Omniscience | 43.3 | -23.6 |
| Artificial Analysis Coding Index | 76.5 | 8.3 |
| critpt | 28.6 | 0 |
| gdpval | 61.9 | 0 |
| GPQA Diamond | 92.6 | 43.4 |
| HLE | 64.5 | 4.8 |
| IFBench | 63.5 | 38 |
| mathvision | 94.8 | 52.4 |
| MMMU-Pro | 84.2 | 44.6 |
| OmniScience Accuracy | 65.3 | 8.2 |
| OmniScience Non-Hallucination | 36.4 | 67.6 |
| scicode | 60.2 | 20.9 |
| TauBench V3 - Banking | 38.1 | 4.5 |
| Terminal-Bench 2.1 | 88 | 0.4 |
| Terminal-Bench Hard | 62.9 | 3 |
| τ²-Bench Telecom (AA run) | 98.5 | 22.2 |
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.