The best ocr models, according to hackers
11 ocr models recommended in Hacker News comments over the past year, ranked by how many separate people recommend each one.
All ocr models
- Mistral OCRPeople recommend Mistral OCR for accurate document parsing with tables, equations and bounding boxes, at low per-page cost. Critics report hallucinated sentences in some versions.69recommend
- PaddleOCRFree open-source OCR toolkit that people use to extract text and layout from documents, running fast on CPU. Critics say it trails some alternatives on accuracy.19recommend
- GLM-OCRFree 0.9 billion parameter OCR model that runs fast on laptops and CPUs with good results on complex documents. Caveat: connection timeouts on large PDFs.16recommend
- DeepSeek-OCRFree open-weights OCR model praised for compressing text into fewer tokens and strong results in use. Critics say newer models now beat it.12recommend
- dots.ocrOpen-source OCR model that people rank among the leading free options, well ahead of traditional OCR. Main caveat: it needs considerable computing resources.8recommend
- Unlimited OCRRecommended for reading very long documents in one pass with a clever sliding-window attention design, and essentially free to run on a decent computer.7recommend
- Amazon TextractRecommended for accurate handwriting recognition and a well-documented Python library; critics call it expensive, prone to surprising costs, and often outperformed by LLMs.6recommend
- Google Cloud Document AIGoogle's cloud service for extracting data from business documents, recommended for OCR quality and speed. People say it works well in simple pipelines but poorly in conversation.6recommend
- PaddleOCR-VLFree open model that people report good results with for document OCR, especially at a small model size.6recommend
- Datalab Chandra OCRPeople recommend Chandra for its accuracy on text and handwriting, and it is open-weight. Critics note weak layout bounding boxes, slow runtime, and restrictive license terms.5recommend
- Surya OCRFree open-source OCR models that people find accurate out of the box and better than rival tools; the main caveat is few settings to tweak.5recommend