| Benchmark | Gemini 3.5 Flash | Qwen3.5 27B |
|---|---|---|
| AA Agentic Index | 70.4 | 54.6 |
| AA Intelligence | 52 | 34.6 |
| AA-LCR | 81 | 72.3 |
| AA-Omniscience | 21.2 | -44 |
| aime_2026 | 95 | 92.6 |
| Artificial Analysis Coding Index | 70.1 | 34.9 |
| charxiv_rq | 84.2 | 79.5 |
| coding_arena_elo | 1490 | 1357 |
| critpt | 13.1 | 0.9 |
| gdpval | 57.8 | 33 |
| GPQA Diamond | 92.2 | 85.8 |
| HLE | 42.7 | 48.5 |
| hmmt_feb_2026 | 95.5 | 84.3 |
| IFBench | 76.3 | 76.5 |
| itbenchSre | 40.3 | 35.5 |
| MCP Atlas | 83.6 | 68.4 |
| MMMU-Pro | 84.3 | 75 |
| OmniScience Accuracy | 51.4 | 20.7 |
| OmniScience Non-Hallucination | 38.2 | 24.9 |
| OSWorld-Verified | 78.4 | 56.2 |
| scicode | 53.1 | 39.5 |
| SWE-bench Pro | 55.1 | 51.2 |
| Terminal-Bench 2.0 | 76.2 | 41.6 |
| Terminal-Bench Hard | 46.2 | 32.6 |
| τ²-Bench Telecom (AA run) | 95.6 | 93.9 |
| τ³-Bench | 25.4 | 68.4 |
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.