Tag
The author claims Qwen3-VL-2B is the only viable vision-language model for JSON extraction on low-end hardware, outperforming larger models like Qwen3-VL-4B, yet it is absent from major benchmarks.