| Benchmark | DeepSeek-V4-Pro | Qwen3.6 27B |
|---|---|---|
| AA Agentic Index | 63.3 | 27.5 |
| AA Intelligence | 53 | 37.7 |
| AA-LCR | 70 | 73.3 |
| AA-Omniscience | -10.6 | -20 |
| Agents' Last Exam | 16.5 | 27.3 |
| aime_2026 | 96.7 | 94.1 |
| Artificial Analysis Coding Index | 59.4 | 53.7 |
| critpt | 13 | 1.1 |
| gdpval | 49 | 32 |
| GPQA Diamond | 90.5 | 87.8 |
| HLE | 48.2 | 24 |
| hmmt_feb_2026 | 95.2 | 84.3 |
| hmmt_nov_2025 | 94.4 | 90.7 |
| IFBench | 76.5 | 69.1 |
| imo_answer_bench | 89.8 | 80.8 |
| livebench_agentic_coding | 42.6 | 39.3 |
| livebench_coding | 70 | 71.8 |
| livebench_data_analysis | 74.5 | 70.4 |
| livebench_instruction_following | 62.4 | 53.2 |
| livebench_language | 78.1 | 63.3 |
| livebench_math | 90.7 | 79.9 |
| livebench_reasoning | 82.7 | 70.3 |
| MMLU-Pro | 87.5 | 86.2 |
| nl2repo | 38.5 | 36.2 |
| OmniScience Accuracy | 43 | 19.6 |
| OmniScience Non-Hallucination | 12.2 | 50.7 |
| scicode | 50 | 39.8 |
| simpleqa | 57.9 | 95.7 |
| SWE-bench Multilingual | 76.2 | 71.3 |
| SWE-bench Pro | 55.4 | 53.5 |
| SWE-bench Verified | 80.6 | 77.2 |
| TauBench V3 - Banking | 30.1 | 16.7 |
| Terminal-Bench 2.0 | 67.9 | 59.3 |
| Terminal-Bench 2.1 | 72.1 | 60.7 |
| Terminal-Bench Hard | 46.2 | 34.8 |
| τ²-Bench Telecom (AA run) | 96.2 | 94.2 |
| τ³-Bench | 25.8 | 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.