| Benchmark | GPT-4o | Qwen3.5 27B |
|---|---|---|
| AA Agentic Index | 9.7 | 54.6 |
| AA Intelligence | 11.2 | 34.6 |
| AA-LCR | 55 | 72.3 |
| AA-Omniscience | -10.5 | -44 |
| AI2D | 94.2 | 92.9 |
| arena_vision | 1162 | 1221 |
| Artificial Analysis Coding Index | 24.2 | 34.9 |
| C-Eval | 76 | 90.5 |
| critpt | 0 | 0.9 |
| ERQA | 35.2 | 60.5 |
| gdpval | 0 | 33 |
| GPQA Diamond | 70.1 | 85.8 |
| HLE | 5.3 | 48.5 |
| IFBench | 36 | 76.5 |
| ifeval | 84.3 | 95 |
| LiveCodeBench v6 | 30.9 | 80.7 |
| longbench_v2 | 48.1 | 60.6 |
| mathvision | 30.4 | 86 |
| MathVista | 63.8 | 87.8 |
| mmlu_prox | 61.1 | 82.2 |
| mmlu_redux | 88 | 93.2 |
| MMLU-Pro | 74.7 | 86.1 |
| mmmlu | 81.4 | 85.9 |
| MMMU | 72.2 | 82.3 |
| MMMU-Pro | 59.9 | 75 |
| MMStar | 63.9 | 81 |
| multichallenge | 40.3 | 60.8 |
| OCRBench | 806 | 89.4 |
| OmniScience Accuracy | 23.7 | 20.7 |
| OmniScience Non-Hallucination | 62.1 | 24.9 |
| OSWorld-Verified | 5 | 56.2 |
| RealWorldQA | 75.4 | 83.7 |
| scicode | 33.4 | 39.5 |
| supergpqa | 42.4 | 65.6 |
| SWE-bench Verified | 38.8 | 75 |
| Terminal-Bench Hard | 8.3 | 32.6 |
| vectara_answer_rate | 93.8 | 99.8 |
| vectara_avg_summary_length | 86.6 | 94.4 |
| vectara_factual_consistency | 90.4 | 87.9 |
| vectara_hallucination_rate ↓ | 9.6 | 12.1 |
| VideoMMMU | 61.2 | 82.3 |
| τ²-Bench Telecom (AA run) | 28.9 | 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.