GLOSSARY
Vision-Language Model (VLM)
A model that reads images as well as text: send a screenshot, chart, photo or diagram and ask questions about it in plain language.
A VLM extends a language model with a vision encoder that turns images into the same kind of tokens the text side consumes. The result is one model that can describe pictures, read text inside screenshots, interpret charts, extract fields from invoices and answer “what is wrong with this UI?” — multimodal input without a separate pipeline.
The capability powers real products now: support tools that read uploaded screenshots, assistants that understand the page you are looking at, and data extraction from scanned documents. Weak spots remain — fine text, precise counting, spatial reasoning and charts with dense axes still trip models up, so outputs on those tasks need the same verification as any extraction.