| Benchmark | GPT-5.4 | Qwen3.5 27B |
|---|---|---|
| AA Agentic Index | 58.2 | 54.6 |
| AA Intelligence | 53.1 | 34.6 |
| AA-LCR | 77.7 | 72.3 |
| AA-Omniscience | 5.8 | -44 |
| aime_2026 | 99.2 | 92.6 |
| arena_vision | 1277 | 1221 |
| Artificial Analysis Coding Index | 71.1 | 34.9 |
| baby_vision | 49.7 | 44.6 |
| browsecomp | 82.7 | 61 |
| charxiv_rq | 82.8 | 79.5 |
| coding_arena_elo | 1457 | 1357 |
| critpt | 23.4 | 0.9 |
| gdpval | 50.1 | 33 |
| GPQA Diamond | 93 | 85.8 |
| HLE | 43.7 | 48.5 |
| hmmt_feb_2026 | 97.7 | 84.3 |
| hmmt_nov_2025 | 95.8 | 89.8 |
| IFBench | 73.9 | 76.5 |
| imo_answer_bench | 91.4 | 79.9 |
| itbenchSre | 34.5 | 35.5 |
| mathvision | 92 | 86 |
| MCP Atlas | 70.6 | 68.4 |
| mcpmark | 62.5 | 36.3 |
| MMLU-Pro | 87.5 | 86.1 |
| MMMU-Pro | 81.2 | 75 |
| nl2repo | 41.3 | 27.3 |
| OmniDocBench 1.5 | 89.1 | 88.9 |
| OmniScience Accuracy | 50.9 | 20.7 |
| OmniScience Non-Hallucination | 17.4 | 24.9 |
| OSWorld-Verified | 75 | 56.2 |
| scicode | 56.6 | 39.5 |
| SWE-bench Multilingual | 71.7 | 69.3 |
| SWE-bench Pro | 59.1 | 51.2 |
| Terminal-Bench 2.0 | 75.1 | 41.6 |
| Terminal-Bench Hard | 57.6 | 32.6 |
| tool_decathlon | 54.6 | 31.5 |
| vectara_answer_rate | 99.9 | 99.8 |
| vectara_avg_summary_length | 81.7 | 94.4 |
| vectara_factual_consistency | 93 | 87.9 |
| vectara_hallucination_rate ↓ | 7 | 12.1 |
| τ²-Bench Telecom (AA run) | 87.1 | 93.9 |
| τ³-Bench | 72.9 | 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.