| Benchmark | GPT-5.5 | Qwen3.5 35B A3B |
|---|---|---|
| AA Agentic Index | 47.4 | 44.1 |
| AA Intelligence | 56.3 | 29.9 |
| AA-LCR | 79 | 68.3 |
| AA-Omniscience | 20.5 | -48.1 |
| aime_2026 | 100 | 93.3 |
| Artificial Analysis Coding Index | 74.9 | 37 |
| browsecomp | 84.4 | 61 |
| coding_arena_elo | 1457 | 1250 |
| critpt | 27.1 | 0.9 |
| gdpval | 49.5 | 20.3 |
| GPQA Diamond | 93.6 | 84.5 |
| HLE | 52.2 | 47.4 |
| hmmt_feb_2026 | 98.5 | 81.8 |
| hmmt_nov_2025 | 96.5 | 89.2 |
| IFBench | 75.9 | 72.5 |
| itbenchSre | 45.8 | 21.5 |
| MCP Atlas | 82.8 | 62.4 |
| MMMU-Pro | 83.2 | 75.1 |
| nl2repo | 50.7 | 20.5 |
| OmniScience Accuracy | 58 | 20.1 |
| OmniScience Non-Hallucination | 12 | 14.6 |
| OSWorld-Verified | 78.7 | 54.5 |
| scicode | 56.1 | 37.7 |
| SWE-bench Pro | 58.6 | 44.6 |
| SWE-bench Verified | 80.6 | 70 |
| TauBench V3 - Banking | 39 | 4.9 |
| Terminal-Bench 2.0 | 82.7 | 40.5 |
| Terminal-Bench 2.1 | 84.3 | 40.8 |
| Terminal-Bench Hard | 60.6 | 26.5 |
| tool_decathlon | 55.6 | 28.7 |
| vectara_answer_rate | 100 | 99.8 |
| vectara_avg_summary_length | 129.6 | 94.9 |
| vectara_factual_consistency | 90.7 | 89.5 |
| vectara_hallucination_rate ↓ | 9.3 | 10.5 |
| τ²-Bench Telecom (AA run) | 93.9 | 89.2 |
| τ³-Bench | 31.3 | 68.9 |
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.