| Benchmark | DeepSeek-R1 | Qwen3.5 397B A17B |
|---|---|---|
| AA Agentic Index | 3.1 | 53.3 |
| AA Intelligence | 18.6 | 34.3 |
| AA-LCR | 56 | 72.7 |
| AA-Omniscience | -31.3 | -30.8 |
| Artificial Analysis Coding Index | 24.6 | 48.2 |
| browsecomp | 8.9 | 69 |
| browsecomp_zh | 35.7 | 70.3 |
| C-Eval | 91.8 | 93 |
| critpt | 0.6 | 1.7 |
| gdpval | 1.5 | 35.8 |
| GPQA Diamond | 81 | 89.3 |
| HLE | 17.7 | 29 |
| HMMT 2025 | 79.4 | 92.7 |
| hmmt_feb_2025 | 76.7 | 94.8 |
| IFBench | 39 | 78.8 |
| ifeval | 83.3 | 92.6 |
| longbench_v2 | 58.3 | 63.2 |
| mmlu_prox | 75.5 | 84.7 |
| mmlu_redux | 93.4 | 94.9 |
| MMLU-Pro | 85 | 87.8 |
| multichallenge | 45 | 67.6 |
| OmniScience Accuracy | 30.7 | 30.8 |
| OmniScience Non-Hallucination | 10.5 | 17.3 |
| scicode | 35.7 | 42 |
| simpleqa_verified | 27.4 | 26 |
| SWE-bench Multilingual | 30.5 | 69.3 |
| SWE-bench Verified | 57.6 | 76.4 |
| TauBench V3 - Banking | 6.4 | 13.4 |
| Terminal-Bench 2.1 | 19.1 | 51.3 |
| Terminal-Bench Hard | 6.1 | 40.9 |
| τ²-Bench Telecom (AA run) | 11.4 | 95.6 |
| τ³-Bench | 6.4 | 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.