Qwen3-VL 8B on a laptop vs Opus 5.5 / Sonnet 5 / GPT-5.6 on 137 messy documents: beat GPT-5.6 on tax forms, lost badly on Indian date formats[R]

A benchmark compared Qwen3-VL 8B Instruct (Q4_K_M) running on a laptop against Claude Opus 5.5, Sonnet 5, and GPT-5.6 Terra using 137 messy documents. Qwen3-VL 8B achieved a 59% success rate, outperforming GPT-5.6 Terra (57%) but falling behind Opus (89%) and Sonnet (85%). While Qwen3-VL 8B beat GPT-5.6 on tax forms, it struggled with Indian date formats. All raw outputs and tests are available on GitHub.