| Benchmark | DeepSeek-V3.2 | Qwen3 VL 4B (Reasoning) |
|---|---|---|
| AA Agentic Index | 39.8 | 14.4 |
| AA Intelligence | 32.8 | 7.7 |
| AA-LCR | 70.7 | 23 |
| AA-Omniscience | -22.5 | -68.4 |
| AIME 2025 | 93.1 | 74.5 |
| Artificial Analysis Coding Index | 44.2 | 6.7 |
| critpt | 2.9 | 0 |
| gdpval | 18.8 | 13.8 |
| GPQA Diamond | 84 | 64.1 |
| HLE | 40.8 | 4.6 |
| HMMT 2025 | 90.2 | 53.1 |
| IFBench | 61 | 36.6 |
| LiveCodeBench | 86 | 51.3 |
| LiveCodeBench v6 | 83.3 | 51.3 |
| mmlu_redux | 93.7 | 86 |
| MMLU-Pro | 86 | 73.6 |
| OmniScience Accuracy | 33 | 12.2 |
| OmniScience Non-Hallucination | 17.3 | 8.2 |
| scicode | 39 | 17.1 |
| Terminal-Bench Hard | 35.6 | 1.5 |
| τ²-Bench Telecom (AA run) | 90.6 | 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.