| Benchmark | Gemini 2.5 Flash | Llama 3.1 Instruct 405B |
|---|---|---|
| AA Agentic Index | 18.8 | 6.3 |
| AA Intelligence | 17.7 | 8.5 |
| AA-LCR | 65.7 | 24.7 |
| AA-Omniscience | -29.8 | -17.1 |
| aider_polyglot | 61.9 | 5.8 |
| AIME 2024 | 82.3 | 23.3 |
| AIR-Bench 2024 | 68.7 | 58.6 |
| anthropic_red_team | 98.8 | 96.5 |
| Artificial Analysis Coding Index | 22.2 | 14.5 |
| bbq | 97.7 | 94.5 |
| critpt | 1.4 | 0 |
| gdpval | 11.9 | 0 |
| GPQA Diamond | 82.8 | 51.5 |
| harmbench | 62.6 | 62.7 |
| HLE | 12.1 | 4.2 |
| IFBench | 50.3 | 39 |
| LiveCodeBench | 76.2 | 30.1 |
| OmniScience Accuracy | 26.1 | 23.2 |
| OmniScience Non-Hallucination | 24.7 | 49 |
| scicode | 35.9 | 29.9 |
| simple_safety_tests | 98 | 98.8 |
| simplebench | 41.2 | 23 |
| simpleqa | 26.9 | 23.2 |
| SWE-bench Verified | 60.4 | 24.5 |
| Terminal-Bench Hard | 13.6 | 6.8 |
| xstest | 98.8 | 95.9 |
| τ²-Bench Telecom (AA run) | 31.6 | 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.