| Benchmark | DeepSeek-V4-Flash | Qwen3.6 35B A3B |
|---|---|---|
| AA Agentic Index | 61.3 | 21.6 |
| AA Intelligence | 51.8 | 32.1 |
| AA-LCR | 74.3 | 66.7 |
| AA-Omniscience | -14.3 | -22.2 |
| aime_2026 | 95.8 | 92.7 |
| Artificial Analysis Coding Index | 69.1 | 41.9 |
| browsecomp | 73.2 | 48.7 |
| critpt | 16.6 | 0.3 |
| gdpval | 52.9 | 27.8 |
| GDPval-AA v2 | 1189 | 1015 |
| GPQA Diamond | 90.8 | 86 |
| HLE | 38.6 | 22.2 |
| hmmt_feb_2026 | 94.8 | 83.6 |
| IFBench | 79.2 | 64.4 |
| imo_answer_bench | 88.4 | 78.9 |
| MCP Atlas | 69 | 62.8 |
| MMLU-Pro | 86.4 | 85.6 |
| nl2repo | 54.2 | 29.4 |
| OmniScience Accuracy | 40.4 | 18.8 |
| OmniScience Non-Hallucination | 15.6 | 49.5 |
| scicode | 49.9 | 35.8 |
| SWE-bench Multilingual | 73.3 | 67.2 |
| SWE-bench Pro | 52.6 | 49.5 |
| SWE-bench Verified | 79 | 73.4 |
| TauBench V3 - Banking | 39.4 | 9.3 |
| Terminal-Bench 2.0 | 56.9 | 51.5 |
| Terminal-Bench 2.1 | 82.7 | 44.9 |
| Terminal-Bench Hard | 38.6 | 34.8 |
| toolathlon | 70.3 | 26.9 |
| τ²-Bench Telecom (AA run) | 95.6 | 95.3 |
| τ³-Bench | 22.9 | 67.2 |
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.