| Benchmark | Gemini 3 Flash Preview | Qwen2.5 Instruct 72B |
|---|---|---|
| AA Intelligence | 38.7 | 10 |
| AA-LCR | 73 | 20.3 |
| AA-Omniscience | 10.1 | -52.2 |
| Artificial Analysis Coding Index | 42.6 | 11.9 |
| critpt | 8.6 | 0 |
| GPQA Diamond | 90.4 | 49.1 |
| harmbench | 70 | 72.8 |
| HLE | 43.5 | 4.2 |
| IFBench | 78 | 36.9 |
| livebench | 72.4 | 52.3 |
| mmmlu | 91.8 | 74.8 |
| OmniScience Accuracy | 53.4 | 17.6 |
| OmniScience Non-Hallucination | 7.6 | 15.3 |
| scicode | 50.6 | 26.7 |
| simpleqa | 68.7 | 10.3 |
| SWE-bench Verified | 78 | 23.8 |
| Terminal-Bench Hard | 38.6 | 4.5 |
| τ²-Bench Telecom (AA run) | 80.4 | 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.