| Benchmark | GPT-5.1 | Qwen3 VL 4B (Reasoning) |
|---|---|---|
| AA Agentic Index | 32.2 | 14.4 |
| AA Intelligence | 37.5 | 7.7 |
| AA-LCR | 76.7 | 23 |
| AA-Omniscience | 5.4 | -68.4 |
| AIME 2025 | 94.2 | 74.5 |
| Artificial Analysis Coding Index | 49.4 | 6.7 |
| critpt | 4.9 | 0 |
| gdpval | 25 | 13.8 |
| GPQA Diamond | 88.1 | 64.1 |
| HLE | 28.5 | 4.6 |
| IFBench | 72.9 | 36.6 |
| LiveCodeBench v6 | 87 | 51.3 |
| MMLU-Pro | 87 | 73.6 |
| MMMU-Pro | 75.5 | 57 |
| OmniScience Accuracy | 37.7 | 12.2 |
| OmniScience Non-Hallucination | 48.1 | 8.2 |
| scicode | 43.3 | 17.1 |
| Terminal-Bench Hard | 45.5 | 1.5 |
| τ²-Bench Telecom (AA run) | 81.9 | 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.