both openai and claude vision models are able do that for me. It is more expensive than tesseract which can run on cpu but I assume it will become similarly cheap in the near future with open models and as AI becomes ubiquitous.