| Benchmark | DeepSeek-V4-Flash | Qwen3.6 27B |
|---|---|---|
| AA Agentic Index | 61.3 | 27.5 |
| AA Intelligence | 51.8 | 37.7 |
| AA-LCR | 74.3 | 73.3 |
| AA-Omniscience | -14.3 | -20 |
| Agents' Last Exam | 25.2 | 27.3 |
| aime_2026 | 95.8 | 94.1 |
| Artificial Analysis Coding Index | 69.1 | 53.7 |
| critpt | 16.6 | 1.1 |
| gdpval | 52.9 | 32 |
| GPQA Diamond | 90.8 | 87.8 |
| HLE | 38.6 | 24 |
| hmmt_feb_2026 | 94.8 | 84.3 |
| IFBench | 79.2 | 69.1 |
| imo_answer_bench | 88.4 | 80.8 |
| livebench_agentic_coding | 46.8 | 39.3 |
| livebench_coding | 75 | 71.8 |
| livebench_data_analysis | 79.3 | 70.4 |
| livebench_instruction_following | 65.5 | 53.2 |
| livebench_language | 79.2 | 63.3 |
| livebench_math | 86.8 | 79.9 |
| livebench_reasoning | 86.6 | 70.3 |
| MMLU-Pro | 86.4 | 86.2 |
| nl2repo | 54.2 | 36.2 |
| OmniScience Accuracy | 40.4 | 19.6 |
| OmniScience Non-Hallucination | 15.6 | 50.7 |
| scicode | 49.9 | 39.8 |
| SWE-bench Multilingual | 73.3 | 71.3 |
| SWE-bench Pro | 52.6 | 53.5 |
| SWE-bench Verified | 79 | 77.2 |
| TauBench V3 - Banking | 39.4 | 16.7 |
| Terminal-Bench 2.0 | 56.9 | 59.3 |
| Terminal-Bench 2.1 | 82.7 | 60.7 |
| Terminal-Bench Hard | 38.6 | 34.8 |
| τ²-Bench Telecom (AA run) | 95.6 | 94.2 |
| τ³-Bench | 22.9 | 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.