Mistral announced OCR 4 via its official news blog, drawing ~288 points / 73 comments on Hacker News alongside Baidu's Unlimited-OCR — a notable same-week clustering of frontier OCR releases. Document parsing remains a core bottleneck for agentic and RAG workflows, and a new first-party Mistral OCR model gives builders another self-hostable/API option to evaluate. Primary source is Mistral's release post; cross-check accuracy claims against your own document set before swapping pipelines.