
Qwen3.5-9B Beats Frontiers on Document Benchmarks
Qwen3.5 models (0.8B to 9B) were evaluated on an open document AI benchmark with 9,000+ real documents. The 9B and 4B variants outperform frontier models like Gemini 3.1 Pro and GPT-5.4 in OCR text extraction, VQA, and KIE tasks. However, they lag significantly in table extraction and handwriting OCR.





