| Benchmark | Gemini 3.6 Flash | Qwen3.6 27B |
|---|---|---|
| AA Agentic Index | 40.5 | 27.5 |
| AA Intelligence | 52 | 37.7 |
| AA-LCR | 79 | 73.3 |
| AA-Omniscience | 22.1 | -20 |
| aime_2026 | 96.7 | 94.1 |
| Artificial Analysis Coding Index | 69.2 | 53.7 |
| charxiv_rq | 85.2 | 78.4 |
| critpt | 10.6 | 1.1 |
| DeepSWE 1.1 | 49 | 13.3 |
| gdpval | 46.1 | 32 |
| GPQA Diamond | 92.8 | 87.8 |
| HLE | 40.8 | 24 |
| hmmt_feb_2026 | 89.4 | 84.3 |
| livebench_agentic_coding | 43.4 | 39.3 |
| livebench_coding | 77.9 | 71.8 |
| livebench_data_analysis | 63 | 70.4 |
| livebench_instruction_following | 75.4 | 53.2 |
| livebench_language | 83.9 | 63.3 |
| livebench_math | 86.4 | 79.9 |
| livebench_reasoning | 85.2 | 70.3 |
| MMMU-Pro | 83.2 | 75.8 |
| OmniScience Accuracy | 50 | 19.6 |
| OmniScience Non-Hallucination | 44.4 | 50.7 |
| OSWorld-Verified | 83 | 63.9 |
| scicode | 52.7 | 39.8 |
| SWE-bench Pro | 58.7 | 53.5 |
| TauBench V3 - Banking | 29.9 | 16.7 |
| Terminal-Bench 2.1 | 78 | 60.7 |
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.