| Benchmark | Qwen2.5 Instruct 72B | Qwen3 VL 4B (Reasoning) |
|---|---|---|
| AA Intelligence | 10 | 7.7 |
| AA-LCR | 20.3 | 23 |
| AA-Omniscience | -52.2 | -68.4 |
| AIME 2024 | 23.3 | 72.9 |
| Artificial Analysis Coding Index | 11.9 | 6.7 |
| critpt | 0 | 0 |
| GPQA Diamond | 49.1 | 64.1 |
| HLE | 4.2 | 4.6 |
| IFBench | 36.9 | 36.6 |
| ifeval | 87.2 | 82.6 |
| LiveCodeBench | 55.5 | 51.3 |
| mmlu_redux | 86.8 | 86 |
| MMLU-Pro | 71.6 | 73.6 |
| OmniScience Accuracy | 17.6 | 12.2 |
| OmniScience Non-Hallucination | 15.3 | 8.2 |
| scicode | 26.7 | 17.1 |
| Terminal-Bench Hard | 4.5 | 1.5 |
| τ²-Bench Telecom (AA run) | 34.5 | 15.5 |
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.