| Benchmark | DeepSeek-V3 | Qwen3.5 397B A17B |
|---|---|---|
| AA Agentic Index | 1.6 | 53.3 |
| AA Intelligence | 15.4 | 34.3 |
| AA-LCR | 41.3 | 72.7 |
| AA-Omniscience | -37.6 | -30.8 |
| Artificial Analysis Coding Index | 23 | 48.2 |
| C-Eval | 90.1 | 93 |
| critpt | 0 | 1.7 |
| gdpval | 0 | 35.8 |
| GPQA Diamond | 68.4 | 89.3 |
| GSM8K | 96.7 | 66.7 |
| HLE | 5.2 | 29 |
| HMMT 2025 | 27.5 | 92.7 |
| hmmt_feb_2025 | 29.2 | 94.8 |
| IFBench | 41 | 78.8 |
| ifeval | 87.3 | 92.6 |
| LiveCodeBench v6 | 46.9 | 83.6 |
| longbench_v2 | 48.7 | 63.2 |
| mmlu_prox | 70.5 | 84.7 |
| mmlu_redux | 90.5 | 94.9 |
| MMLU-Pro | 81.2 | 87.8 |
| mmmlu | 79.4 | 88.5 |
| multichallenge | 31.4 | 67.6 |
| OmniScience Accuracy | 25.4 | 30.8 |
| OmniScience Non-Hallucination | 23.3 | 17.3 |
| scicode | 35.8 | 42 |
| supergpqa | 53.7 | 70.4 |
| SWE-bench Verified | 42 | 76.4 |
| TauBench V3 - Banking | 4.7 | 13.4 |
| Terminal-Bench 2.1 | 16.9 | 51.3 |
| Terminal-Bench Hard | 15.2 | 40.9 |
| τ²-Bench Telecom (AA run) | 47.1 | 95.6 |
| τ³-Bench | 4.7 | 13.4 |
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.