| Benchmark | o4-mini | Qwen3.5 27B |
|---|---|---|
| AA Agentic Index | 36.1 | 54.6 |
| AA Intelligence | 26.1 | 34.6 |
| AA-LCR | 60 | 72.3 |
| AA-Omniscience | -35.7 | -44 |
| arena_vision | 1201 | 1221 |
| Artificial Analysis Coding Index | 25.6 | 34.9 |
| browsecomp | 51.5 | 61 |
| critpt | 0.6 | 0.9 |
| gdpval | 25.4 | 33 |
| GPQA Diamond | 81.4 | 85.8 |
| HLE | 16.5 | 48.5 |
| IFBench | 68.7 | 76.5 |
| MathVista | 84.3 | 87.8 |
| mmlu_prox | 69.3 | 82.2 |
| MMMU | 81.6 | 82.3 |
| MMMU-Pro | 69.2 | 75 |
| multichallenge | 44.9 | 60.8 |
| OmniScience Accuracy | 24.8 | 20.7 |
| OmniScience Non-Hallucination | 19.5 | 24.9 |
| scicode | 46.5 | 39.5 |
| SWE-bench Verified | 68.1 | 75 |
| Terminal-Bench Hard | 15.2 | 32.6 |
| vectara_answer_rate | 99.2 | 99.8 |
| vectara_avg_summary_length | 130.9 | 94.4 |
| vectara_factual_consistency | 81.4 | 87.9 |
| vectara_hallucination_rate ↓ | 18.6 | 12.1 |
| τ²-Bench Telecom (AA run) | 55.6 | 93.9 |
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.