| Benchmark | GPT-5.2 | Qwen3.6 27B |
|---|---|---|
| AA Agentic Index | 60.2 | 27.5 |
| AA Intelligence | 43.3 | 37.7 |
| AA-LCR | 79.3 | 73.3 |
| AA-Omniscience | 0.3 | -20 |
| aime_2026 | 98.3 | 94.1 |
| Artificial Analysis Coding Index | 48.7 | 53.7 |
| charxiv_rq | 82.1 | 78.4 |
| critpt | 11.6 | 1.1 |
| gdpval | 48.3 | 32 |
| GPQA Diamond | 92.4 | 87.8 |
| HLE | 45.5 | 24 |
| HMMT 2025 | 99.4 | 90.7 |
| hmmt_feb_2025 | 100 | 93.8 |
| hmmt_feb_2026 | 97 | 84.3 |
| hmmt_nov_2025 | 99.2 | 90.7 |
| IFBench | 75.4 | 69.1 |
| imo_answer_bench | 86.3 | 80.8 |
| livebench_agentic_coding | 50.3 | 39.3 |
| livebench_coding | 76.1 | 71.8 |
| livebench_data_analysis | 78.2 | 70.4 |
| livebench_instruction_following | 61.8 | 53.2 |
| livebench_language | 79.8 | 63.3 |
| livebench_math | 93.2 | 79.9 |
| livebench_reasoning | 83.2 | 70.3 |
| mathvision | 83 | 85.1 |
| MathVista | 82.8 | 87.4 |
| MMLU-Pro | 87 | 86.2 |
| MMMU-Pro | 79.5 | 75.8 |
| OCRBench | 80.7 | 89.4 |
| OmniDocBench 1.5 | 85.7 | 89.4 |
| OmniScience Accuracy | 44.3 | 19.6 |
| OmniScience Non-Hallucination | 38.4 | 50.7 |
| scicode | 52.1 | 39.8 |
| SimpleVQA | 55.8 | 56.1 |
| SWE-bench Multilingual | 72 | 71.3 |
| SWE-bench Pro | 55.6 | 53.5 |
| SWE-bench Verified | 80 | 77.2 |
| TauBench V3 - Banking | 11.1 | 16.7 |
| Terminal-Bench 2.0 | 54 | 59.3 |
| Terminal-Bench Hard | 62.2 | 34.8 |
| VideoMMMU | 85.9 | 84.4 |
| τ²-Bench Telecom (AA run) | 98.7 | 94.2 |
| τ³-Bench | 12 | 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.