| Benchmark | Gemma 4 31B | Llama 3.1 Instruct 405B |
|---|---|---|
| AA Agentic Index | 14.4 | 6.3 |
| AA Intelligence | 29.7 | 8.5 |
| AA-LCR | 68.3 | 24.7 |
| AA-Omniscience | -47.9 | -17.1 |
| Artificial Analysis Coding Index | 43.4 | 14.5 |
| C-Eval | 82.6 | 72.5 |
| critpt | 1.4 | 0 |
| gdpval | 15.5 | 0 |
| GPQA Diamond | 85.7 | 51.5 |
| HLE | 26.5 | 4.2 |
| IFBench | 75.6 | 39 |
| mmlu_redux | 93.7 | 86.2 |
| MMLU-Pro | 85.2 | 73.4 |
| mmmlu | 88.4 | 73.8 |
| OmniScience Accuracy | 20 | 23.2 |
| OmniScience Non-Hallucination | 18.1 | 49 |
| scicode | 43.4 | 29.9 |
| SWE-bench Pro | 35.7 | 11.2 |
| SWE-bench Verified | 52 | 24.5 |
| Terminal-Bench Hard | 36.4 | 6.8 |
| τ²-Bench Telecom (AA run) | 65.5 | 19 |
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.