| Benchmark | Gemini 3 Flash Preview | Qwen3.5 27B |
|---|---|---|
| AA Agentic Index | 49.7 | 54.6 |
| AA Intelligence | 38.7 | 34.6 |
| AA-LCR | 73 | 72.3 |
| AA-Omniscience | 10.1 | -44 |
| aime_2026 | 96.7 | 92.6 |
| arena_vision | 1260 | 1221 |
| Artificial Analysis Coding Index | 42.6 | 34.9 |
| charxiv_rq | 80.3 | 79.5 |
| coding_arena_elo | 1387 | 1357 |
| critpt | 8.6 | 0.9 |
| gdpval | 35.2 | 33 |
| GPQA Diamond | 90.4 | 85.8 |
| HLE | 43.5 | 48.5 |
| hmmt_feb_2025 | 97.5 | 92 |
| hmmt_feb_2026 | 89.4 | 84.3 |
| hmmt_nov_2025 | 93.3 | 89.8 |
| IFBench | 78 | 76.5 |
| MCP Atlas | 62 | 68.4 |
| mmmlu | 91.8 | 85.9 |
| MMMU-Pro | 81.2 | 75 |
| OmniDocBench 1.5 | 12.1 | 88.9 |
| OmniScience Accuracy | 53.4 | 20.7 |
| OmniScience Non-Hallucination | 7.6 | 24.9 |
| OSWorld-Verified | 65.1 | 56.2 |
| scicode | 50.6 | 39.5 |
| screenspot_pro_no_tools | 69.1 | 70.3 |
| SWE-bench Multilingual | 72.7 | 69.3 |
| SWE-bench Pro | 49.6 | 51.2 |
| SWE-bench Verified | 78 | 75 |
| Terminal-Bench 2.0 | 47.6 | 41.6 |
| Terminal-Bench Hard | 38.6 | 32.6 |
| vectara_answer_rate | 99.8 | 99.8 |
| vectara_avg_summary_length | 90.2 | 94.4 |
| vectara_factual_consistency | 86.5 | 87.9 |
| vectara_hallucination_rate ↓ | 13.5 | 12.1 |
| VideoMMMU | 86.9 | 82.3 |
| τ²-Bench Telecom (AA run) | 80.4 | 93.9 |
| τ³-Bench | 17.5 | 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.