| Benchmark | DeepSeek-R1 | Qwen3 VL 4B (Reasoning) |
|---|---|---|
| AA Agentic Index | 3.1 | 14.4 |
| AA Intelligence | 18.6 | 7.7 |
| AA-LCR | 56 | 23 |
| AA-Omniscience | -31.3 | -68.4 |
| AIME 2024 | 91.4 | 72.9 |
| AIME 2025 | 87.5 | 74.5 |
| Artificial Analysis Coding Index | 24.6 | 6.7 |
| critpt | 0.6 | 0 |
| gdpval | 1.5 | 13.8 |
| GPQA Diamond | 81 | 64.1 |
| HLE | 17.7 | 4.6 |
| HMMT 2025 | 79.4 | 53.1 |
| IFBench | 39 | 36.6 |
| ifeval | 83.3 | 82.6 |
| LiveCodeBench | 84.4 | 51.3 |
| mmlu_prox | 75.5 | 65 |
| mmlu_redux | 93.4 | 86 |
| MMLU-Pro | 85 | 73.6 |
| OmniScience Accuracy | 30.7 | 12.2 |
| OmniScience Non-Hallucination | 10.5 | 8.2 |
| scicode | 35.7 | 17.1 |
| Terminal-Bench Hard | 6.1 | 1.5 |
| τ²-Bench Telecom (AA run) | 11.4 | 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.