| Benchmark | GPT-5.4 | Qwen3.6 35B A3B |
|---|---|---|
| AA Agentic Index | 58.2 | 21.6 |
| AA Intelligence | 53.1 | 32.1 |
| AA-LCR | 77.7 | 66.7 |
| AA-Omniscience | 5.8 | -22.2 |
| aime_2026 | 99.2 | 92.7 |
| Artificial Analysis Coding Index | 71.1 | 41.9 |
| browsecomp | 82.7 | 48.7 |
| charxiv_rq | 82.8 | 78 |
| critpt | 23.4 | 0.3 |
| gdpval | 50.1 | 27.8 |
| GPQA Diamond | 93 | 86 |
| HLE | 43.7 | 22.2 |
| hmmt_feb_2026 | 97.7 | 83.6 |
| hmmt_nov_2025 | 95.8 | 89.1 |
| IFBench | 73.9 | 64.4 |
| imo_answer_bench | 91.4 | 78.9 |
| MCP Atlas | 70.6 | 62.8 |
| mcpmark | 62.5 | 37 |
| MMLU-Pro | 87.5 | 85.6 |
| MMMU-Pro | 81.2 | 75.3 |
| nl2repo | 41.3 | 29.4 |
| OmniDocBench 1.5 | 89.1 | 89.9 |
| OmniScience Accuracy | 50.9 | 18.8 |
| OmniScience Non-Hallucination | 17.4 | 49.5 |
| scicode | 56.6 | 35.8 |
| SWE-bench Multilingual | 71.7 | 67.2 |
| SWE-bench Pro | 59.1 | 49.5 |
| TauBench V3 - Banking | 39.6 | 9.3 |
| Terminal-Bench 2.0 | 75.1 | 51.5 |
| Terminal-Bench 2.1 | 78.3 | 44.9 |
| Terminal-Bench Hard | 57.6 | 34.8 |
| tool_decathlon | 54.6 | 26.9 |
| toolathlon | 54.6 | 26.9 |
| τ²-Bench Telecom (AA run) | 87.1 | 95.3 |
| τ³-Bench | 72.9 | 67.2 |
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.