| Benchmark | DeepSeek-V3.2 | Qwen3.6 27B |
|---|---|---|
| AA Agentic Index | 39.8 | 27.5 |
| AA Intelligence | 32.8 | 37.7 |
| AA-LCR | 70.7 | 73.3 |
| AA-Omniscience | -22.5 | -20 |
| aime_2026 | 95.1 | 94.1 |
| Artificial Analysis Coding Index | 44.2 | 53.7 |
| critpt | 2.9 | 1.1 |
| gdpval | 18.8 | 32 |
| GPQA Diamond | 84 | 87.8 |
| HLE | 40.8 | 24 |
| HMMT 2025 | 90.2 | 90.7 |
| hmmt_feb_2025 | 92.5 | 93.8 |
| hmmt_feb_2026 | 84.1 | 84.3 |
| hmmt_nov_2025 | 90.2 | 90.7 |
| IFBench | 61 | 69.1 |
| imo_answer_bench | 78.3 | 80.8 |
| LiveCodeBench v6 | 83.3 | 83.9 |
| mmlu_redux | 93.7 | 93.5 |
| MMLU-Pro | 86 | 86.2 |
| OmniScience Accuracy | 33 | 19.6 |
| OmniScience Non-Hallucination | 17.3 | 50.7 |
| scicode | 39 | 39.8 |
| SWE-bench Multilingual | 70.2 | 71.3 |
| SWE-bench Pro | 15.6 | 53.5 |
| SWE-bench Verified | 73.1 | 77.2 |
| TauBench V3 - Banking | 18.8 | 16.7 |
| Terminal-Bench 2.0 | 46.4 | 59.3 |
| Terminal-Bench 2.1 | 46.8 | 60.7 |
| Terminal-Bench Hard | 35.6 | 34.8 |
| τ²-Bench Telecom (AA run) | 90.6 | 94.2 |
| τ³-Bench | 69.2 | 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.