December 19, 2025
OCR wars: baselines and bruised egos
Mistral OCR 3
Mistral OCR 3 drops with cheap pricing—commenters cry 'weak baselines' and Google still king
TLDR: Mistral launched OCR 3, a budget tool to read text from images and PDFs, claiming top accuracy and a drag‑and‑drop playground. Commenters are split: many slam “weak baselines” and say open‑source and Google beat it, making the launch a price play with credibility questions for those needing dependable document reading.
Mistral’s new “read text from images” tool, OCR 3, landed with big promises: cleaner tables, better handwriting, and a drag‑and‑drop Document AI Playground. They’re flexing price too—about $2 per 1,000 pages (and roughly $1 in batch). But the internet’s reaction? Spicy. The loudest chorus accuses Mistral of cherry-picking easy comparisons. One viral quip linked a tweet calling for “stronger baselines” like Chandra, MonkeyOCR, and MinerU, basically saying: don’t brag until you beat the best (tweet). Others waved a scoreboard: a community leaderboard suggests open source tools like PaddleOCR and MinerU might already outshine Mistral (codesota.com/ocr).
Then came the regional drama: Is Mistral the EU’s great AI hope or just chasing “the fringes” while the US giants dominate? Skeptics say Google still leads OCR, with Mistral fighting for second. Some asked if OpenRouter still funnels OCR work to Mistral—cue theories about platform politics. The memes? “Benchmarking against my toaster,” “OCR Hunger Games,” and “EU boss fight unlocked.”
To be fair, fans like the price and the focus on handwritten messes. But the crowd’s mood is clear: nice features, questionable flex. Until Mistral shows head‑to‑head wins against the open‑source all‑stars, the comments section isn’t buying the “state‑of‑the‑art” swagger.
Key Points
- •Mistral AI launched Mistral OCR 3, a new OCR model for accurate, efficient document processing.
- •The model claims a 74% overall win rate over Mistral OCR 2 on forms, scans, tables, and handwriting.
- •Outputs support markdown with HTML-based table reconstruction to preserve document structure.
- •Mistral OCR 3 powers the Document AI Playground in Mistral AI Studio for PDF/image parsing.
- •Pricing is $2 per 1,000 pages, with a 50% Batch API discount to $1 per 1,000 pages; API model ID is mistral-ocr-2512.