This feature works with OpenAI, Anthropic, Gemini, Mistral and Ollama via various plugins.

The new Gemini 2.5 Pro is particularly good at this - especially with gnarly sources like PDFs of scanned documents.

I just upgraded the llm-gemini plugin to support the paid tier of Gemini 2.5 Pro which was released this morning: https://simonwillison.net/2025/Apr/4/gemini-25-pro-pricing/