| Benchmark | DeepSeek-V3 | Qwen3.6 27B |
|---|---|---|
| AA Agentic Index | 1.6 | 27.5 |
| AA Intelligence | 15.4 | 37.7 |
| AA-LCR | 41.3 | 73.3 |
| AA-Omniscience | -37.6 | -20 |
| Artificial Analysis Coding Index | 23 | 53.7 |
| C-Eval | 90.1 | 91.4 |
| critpt | 0 | 1.1 |
| gdpval | 0 | 32 |
| GPQA Diamond | 68.4 | 87.8 |
| hellaswag | 88.9 | 90 |
| HLE | 5.2 | 24 |
| HMMT 2025 | 27.5 | 90.7 |
| hmmt_feb_2025 | 29.2 | 93.8 |
| humaneval | 92.1 | 78.2 |
| IFBench | 41 | 69.1 |
| LiveCodeBench v6 | 46.9 | 83.9 |
| mmlu_redux | 90.5 | 93.5 |
| MMLU-Pro | 81.2 | 86.2 |
| OmniScience Accuracy | 25.4 | 19.6 |
| OmniScience Non-Hallucination | 23.3 | 50.7 |
| scicode | 35.8 | 39.8 |
| simpleqa | 27.7 | 95.7 |
| supergpqa | 53.7 | 66 |
| SWE-bench Verified | 42 | 77.2 |
| TauBench V3 - Banking | 4.7 | 16.7 |
| Terminal-Bench 2.1 | 16.9 | 60.7 |
| Terminal-Bench Hard | 15.2 | 34.8 |
| τ²-Bench Telecom (AA run) | 47.1 | 94.2 |
| τ³-Bench | 4.7 | 15.3 |
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.