| Benchmark | Gemini 2.5 Flash | Qwen2.5 Instruct 72B |
|---|---|---|
| AA Intelligence | 17.7 | 10 |
| AA-LCR | 65.7 | 20.3 |
| AA-Omniscience | -29.8 | -52.2 |
| aider_polyglot | 61.9 | 7.6 |
| AIME 2024 | 82.3 | 23.3 |
| AIR-Bench 2024 | 68.7 | 59 |
| anthropic_red_team | 98.8 | 99.6 |
| Artificial Analysis Coding Index | 22.2 | 11.9 |
| bbq | 97.7 | 95.4 |
| critpt | 1.4 | 0 |
| GPQA Diamond | 82.8 | 49.1 |
| harmbench | 62.6 | 72.8 |
| HLE | 12.1 | 4.2 |
| IFBench | 50.3 | 36.9 |
| LiveCodeBench | 76.2 | 55.5 |
| OmniScience Accuracy | 26.1 | 17.6 |
| OmniScience Non-Hallucination | 24.7 | 15.3 |
| scicode | 35.9 | 26.7 |
| simple_safety_tests | 98 | 100 |
| simpleqa | 26.9 | 10.3 |
| SWE-bench Verified | 60.4 | 23.8 |
| Terminal-Bench Hard | 13.6 | 4.5 |
| xstest | 98.8 | 97.9 |
| τ²-Bench Telecom (AA run) | 31.6 | 34.5 |
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.