| Benchmark | DeepSeek-R1 | Qwen3.5 35B A3B |
|---|---|---|
| AA Agentic Index | 3.1 | 44.1 |
| AA Intelligence | 18.6 | 29.9 |
| AA-LCR | 56 | 68.3 |
| AA-Omniscience | -31.3 | -48.1 |
| Artificial Analysis Coding Index | 24.6 | 37 |
| browsecomp | 8.9 | 61 |
| browsecomp_zh | 35.7 | 69.5 |
| C-Eval | 91.8 | 90.2 |
| critpt | 0.6 | 0.9 |
| gdpval | 1.5 | 20.3 |
| GPQA Diamond | 81 | 84.5 |
| HLE | 17.7 | 47.4 |
| HMMT 2025 | 79.4 | 89.2 |
| hmmt_feb_2025 | 76.7 | 89 |
| IFBench | 39 | 72.5 |
| ifeval | 83.3 | 91.9 |
| longbench_v2 | 58.3 | 59 |
| mmlu | 92.9 | 81.1 |
| mmlu_prox | 75.5 | 81 |
| mmlu_redux | 93.4 | 93.3 |
| MMLU-Pro | 85 | 85.3 |
| multichallenge | 45 | 60 |
| OmniScience Accuracy | 30.7 | 20.1 |
| OmniScience Non-Hallucination | 10.5 | 14.6 |
| scicode | 35.7 | 37.7 |
| SWE-bench Multilingual | 30.5 | 60.3 |
| SWE-bench Verified | 57.6 | 70 |
| TauBench V3 - Banking | 6.4 | 4.9 |
| Terminal-Bench 2.1 | 19.1 | 40.8 |
| Terminal-Bench Hard | 6.1 | 26.5 |
| vectara_answer_rate | 97 | 99.8 |
| vectara_avg_summary_length | 93.5 | 94.9 |
| vectara_factual_consistency | 88.7 | 89.5 |
| vectara_hallucination_rate ↓ | 11.3 | 10.5 |
| τ²-Bench Telecom (AA run) | 11.4 | 89.2 |
| τ³-Bench | 6.4 | 68.9 |
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.