| Benchmark | Gemini 3 Flash Preview | Llama 3.1 Instruct 405B |
|---|---|---|
| AA Agentic Index | 49.7 | 6.3 |
| AA Intelligence | 38.7 | 8.5 |
| AA-LCR | 73 | 24.7 |
| AA-Omniscience | 10.1 | -17.1 |
| Artificial Analysis Coding Index | 42.6 | 14.5 |
| critpt | 8.6 | 0 |
| gdpval | 35.2 | 0 |
| GPQA Diamond | 90.4 | 51.5 |
| harmbench | 70 | 62.7 |
| HLE | 43.5 | 4.2 |
| IFBench | 78 | 39 |
| mmmlu | 91.8 | 73.8 |
| OmniScience Accuracy | 53.4 | 23.2 |
| OmniScience Non-Hallucination | 7.6 | 49 |
| scicode | 50.6 | 29.9 |
| simplebench | 61.1 | 23 |
| simpleqa | 68.7 | 23.2 |
| SWE-bench Pro | 49.6 | 11.2 |
| SWE-bench Verified | 78 | 24.5 |
| Terminal-Bench Hard | 38.6 | 6.8 |
| τ²-Bench Telecom (AA run) | 80.4 | 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.