| Benchmark | GPT-5.6 Sol | Qwen3.5 27B |
|---|---|---|
| AA Agentic Index | 57.8 | 54.6 |
| AA Intelligence | 61 | 34.6 |
| AA-LCR | 77.7 | 72.3 |
| AA-Omniscience | 22 | -44 |
| aime_2026 | 99.9 | 92.6 |
| Artificial Analysis Coding Index | 78.3 | 34.9 |
| browsecomp | 90.4 | 61 |
| charxiv_rq | 87.8 | 79.5 |
| coding_arena_elo | 1619 | 1357 |
| critpt | 32.3 | 0.9 |
| gdpval | 61.1 | 33 |
| GPQA Diamond | 94.6 | 85.8 |
| HLE | 49.5 | 48.5 |
| IFBench | 72.7 | 76.5 |
| itbenchSre | 56.2 | 35.5 |
| longbench_v2 | 67.1 | 60.6 |
| mathvision | 95.8 | 86 |
| MCP Atlas | 83.6 | 68.4 |
| MMMU-Pro | 84.6 | 75 |
| MMVU | 81.2 | 73.3 |
| OmniScience Accuracy | 59.4 | 20.7 |
| OmniScience Non-Hallucination | 10.6 | 24.9 |
| OSWorld-Verified | 83 | 56.2 |
| scicode | 56.9 | 39.5 |
| SWE-bench Pro | 64.6 | 51.2 |
| Terminal-Bench Hard | 65.9 | 32.6 |
| τ²-Bench Telecom (AA run) | 85.1 | 93.9 |
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.