Mistral OCR 3 extracts text and images from documents with SOTA accuracy. It handles handwriting, complex tables, and scanned forms, outputting clean markdown.
Hi everyone!
Just checked out Mistral OCR 3. The handwriting recognition capabilities look stunning, handling even messy scribbles that usually break other models.
The performance on complex tables and scanned forms also seems very robust. On the technical side, getting clean Markdown with HTML table structures straight from the API is a huge win for RAG pipelines.
At $1/1k pages via Batch API, the pricing is also very competitive.
It's good OCR model but they have increased prices by 100% for this. If they keep doing that, this will not be really positive for their business and they will lose a lot of customers even though their model is good
Wow, this is great, we should look if this is suitable for our DeepTagger,
DeepTagger is also about extracting structured data, but it's easier and more visual to create JSON schema, and it's easier to inspect result, as it's shown right on the document.
Open weights + OCR 3 looks handy. If it really handles handwriting and gnarly tables, that could save me hours on invoices. Clean markdown is a nice touch. How’s it on noisy scans / mixed languages? I might finally retire my hacky Tesseract + scripts.
Report
Congrats on the launch! I just tried OCR 3 and decided to push it a bit by uploading an image with non-English text. It handled it flawlessly—nice work. Also, the subscription flow was smooth and frictionless. I really appreciate how easy it is to use. I’ll definitely keep an eye on the product and continue using it.
Report
I think this is awesome and can be game-changing for note-taking apps. Also, for people with ADHD, which takes benefit more from handwriting than typing.
I am personally developing an app that could greatly benefit from such an update.
Report
Love this positioning. Open models with permissive licensing build real trust, and having optimized commercial versions alongside makes adoption practical at scale.
Clear separation between community-first openness and production-ready performance — that balance matters.
Report
It’s refreshing to see products in the market that focus on solving real problems instead of forcing AI into everything. We’re currently exploring an OCR solution for one of our Mutual Funds products to scan international documents. Given the highly confidential nature of these customer documents, we need a solution that does not expose sensitive data to external or generative AI systems. Is this something your team could support?
Report
Congrats on OCR 3 — getting clean Markdown + proper tables straight from the API is huge for RAG pipelines. Quick Q: how does it handle mixed-language docs + noisy/rotated scans, and do you expose confidence scores / bounding boxes?
Replies
Flowtica Scribe
Camocopy
It's good OCR model but they have increased prices by 100% for this. If they keep doing that, this will not be really positive for their business and they will lose a lot of customers even though their model is good
DeepTagger
Wow, this is great, we should look if this is suitable for our DeepTagger,
DeepTagger is also about extracting structured data, but it's easier and more visual to create JSON schema, and it's easier to inspect result, as it's shown right on the document.
Thos benchmarks are truly impressive! 🚀
Product Hunt Wrapped 2025
Open weights + OCR 3 looks handy. If it really handles handwriting and gnarly tables, that could save me hours on invoices. Clean markdown is a nice touch. How’s it on noisy scans / mixed languages? I might finally retire my hacky Tesseract + scripts.
Congrats on the launch!
I just tried OCR 3 and decided to push it a bit by uploading an image with non-English text. It handled it flawlessly—nice work.
Also, the subscription flow was smooth and frictionless. I really appreciate how easy it is to use.
I’ll definitely keep an eye on the product and continue using it.
I think this is awesome and can be game-changing for note-taking apps. Also, for people with ADHD, which takes benefit more from handwriting than typing.
I am personally developing an app that could greatly benefit from such an update.
Love this positioning. Open models with permissive licensing build real trust, and having optimized commercial versions alongside makes adoption practical at scale.
Clear separation between community-first openness and production-ready performance — that balance matters.
Congrats on OCR 3 — getting clean Markdown + proper tables straight from the API is huge for RAG pipelines. Quick Q: how does it handle mixed-language docs + noisy/rotated scans, and do you expose confidence scores / bounding boxes?