| Benchmark | GPT-5 | Qwen3 VL 4B (Reasoning) |
|---|---|---|
| AA Agentic Index | 49.7 | 14.4 |
| AA Intelligence | 35.3 | 7.7 |
| AA-LCR | 76.3 | 23 |
| AA-Omniscience | -8.7 | -68.4 |
| AIME 2025 | 95 | 74.5 |
| Artificial Analysis Coding Index | 38.9 | 6.7 |
| critpt | 5.7 | 0 |
| ERQA | 65.7 | 47.3 |
| gdpval | 32.5 | 13.8 |
| GPQA Diamond | 87.3 | 64.1 |
| HLE | 28.5 | 4.6 |
| HMMT 2025 | 93.3 | 53.1 |
| IFBench | 73.1 | 36.6 |
| LiveCodeBench | 85 | 51.3 |
| LiveCodeBench v6 | 87 | 51.3 |
| mmlu_redux | 95.3 | 86 |
| MMLU-Pro | 87.5 | 73.6 |
| MMMU-Pro | 78.4 | 57 |
| OmniScience Accuracy | 40.3 | 12.2 |
| OmniScience Non-Hallucination | 21 | 8.2 |
| scicode | 43 | 17.1 |
| Terminal-Bench Hard | 37.9 | 1.5 |
| VideoMMMU | 84.6 | 69.4 |
| τ²-Bench Telecom (AA run) | 86.5 | 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.