| Benchmark | DeepSeek-V4-Pro | Qwen3 VL 4B (Reasoning) |
|---|---|---|
| AA Agentic Index | 63.3 | 14.4 |
| AA Intelligence | 53 | 7.7 |
| AA-LCR | 70 | 23 |
| AA-Omniscience | -10.6 | -68.4 |
| Artificial Analysis Coding Index | 59.4 | 6.7 |
| critpt | 13 | 0 |
| gdpval | 49 | 13.8 |
| GPQA Diamond | 90.5 | 64.1 |
| HLE | 48.2 | 4.6 |
| IFBench | 76.5 | 36.6 |
| LiveCodeBench | 93.5 | 51.3 |
| MMLU-Pro | 87.5 | 73.6 |
| OmniScience Accuracy | 43 | 12.2 |
| OmniScience Non-Hallucination | 12.2 | 8.2 |
| scicode | 50 | 17.1 |
| Terminal-Bench Hard | 46.2 | 1.5 |
| τ²-Bench Telecom (AA run) | 96.2 | 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.