Try a picture with a known answer
Give a vision model a picture and a question. In this example, Qwen3-VL 8B looks at a test card and describes what it sees. It is a larger download than the first text model. You need both the model and its matching image projection file. From the repository root, use the included test card:
version/v8/scripts/cks-v8-run run \
hf://Qwen/Qwen3-VL-8B-Instruct-GGUF/Qwen3VL-8B-Instruct-Q4_K_M.gguf \
--mmproj hf://Qwen/Qwen3-VL-8B-Instruct-GGUF/mmproj-Qwen3VL-8B-Instruct-Q8_0.gguf \
--image-path version/v8/test_assets/v8_vision_doc_card_72.ppm \
--prompt "Explain this image." \
--context-len 1024 \
--force-convert --force-compile \
--thinking-mode suppressedOpen the input image and check whether the description matches what you see. For a receipt, check the actual fields or totals; a plausible caption alone cannot establish OCR accuracy.
Other vision options
Qwen3.6-VL has a separate image evaluation path. Cohere North Micro Vision 2.4B has produced a field answer from a real image, with broader task evaluation in progress. Gemma4 E4B has an initial image-input check. For Kimi-VL A3B and Muse-Glimmer 30B, the published results cited in the model catalog cover text. Vision and OCR commands