| Benchmark | GPT-4o | Qwen3.5 35B A3B |
|---|---|---|
| AA Agentic Index | 9.7 | 44.1 |
| AA Intelligence | 11.2 | 29.9 |
| AA-LCR | 55 | 68.3 |
| AA-Omniscience | -10.5 | -48.1 |
| AI2D | 94.2 | 92.6 |
| Artificial Analysis Coding Index | 24.2 | 37 |
| C-Eval | 76 | 90.2 |
| critpt | 0 | 0.9 |
| ERQA | 35.2 | 64.8 |
| gdpval | 0 | 20.3 |
| GPQA Diamond | 70.1 | 84.5 |
| GSM8K | 95.6 | 90.1 |
| HLE | 5.3 | 47.4 |
| humaneval | 90.6 | 66.5 |
| IFBench | 36 | 72.5 |
| ifeval | 84.3 | 91.9 |
| LiveCodeBench v6 | 30.9 | 74.6 |
| longbench_v2 | 48.1 | 59 |
| mathvision | 30.4 | 83.9 |
| MathVista | 63.8 | 86.2 |
| mmlu | 88.1 | 81.1 |
| mmlu_prox | 61.1 | 81 |
| mmlu_redux | 88 | 93.3 |
| MMLU-Pro | 74.7 | 85.3 |
| mmmlu | 81.4 | 85.2 |
| MMMU | 72.2 | 81.4 |
| MMMU-Pro | 59.9 | 75.1 |
| MMStar | 63.9 | 81.9 |
| multichallenge | 40.3 | 60 |
| OCRBench | 806 | 91 |
| OmniScience Accuracy | 23.7 | 20.1 |
| OmniScience Non-Hallucination | 62.1 | 14.6 |
| OpenBookQA | 96.8 | 44.2 |
| OSWorld-Verified | 5 | 54.5 |
| RealWorldQA | 75.4 | 84.1 |
| scicode | 33.4 | 37.7 |
| supergpqa | 42.4 | 63.4 |
| SWE-bench Verified | 38.8 | 70 |
| Terminal-Bench Hard | 8.3 | 26.5 |
| vectara_answer_rate | 93.8 | 99.8 |
| vectara_avg_summary_length | 86.6 | 94.9 |
| vectara_factual_consistency | 90.4 | 89.5 |
| vectara_hallucination_rate ↓ | 9.6 | 10.5 |
| VideoMMMU | 61.2 | 80.4 |
| τ²-Bench Telecom (AA run) | 28.9 | 89.2 |
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.