{"slug": "deepseek-ocr", "name": "DeepSeek-OCR", "category": "ai", "category_name": "AI tools and apps", "description": "", "recommenders": 6, "commenters": 9, "warners": 2, "score": 3.955, "rising": 0.69, "first": "2025-11-01", "latest": "2026-10-25", "why": "DeepSeek-OCR proved 10x+ compression from visual embedding of text", "why_is_quote": true, "has_verdict": false, "sentiment": {"recommend": 6, "neutral": 1, "compare": 0, "warn": 2}, "spark": [1, 0, 0, 0, 0, 0, 0, 1, 2, 1, 0, 1], "listed": true, "loved": [], "caveat": "", "short_quote": "DeepSeek-OCR proved 10x+ compression from visual embedding of text", "kind": "", "icon": true, "domain": null, "github_repo": "github.com/deepseek-ai/deepseek-ocr", "official_url": "https://github.com/deepseek-ai/DeepSeek-OCR", "pricing": "", "aliases": ["DeepSeek-OCR", "DeepSeek OCR"], "verdict": null, "praise": [{"label": "", "summary": "", "quotes": [{"id": 803737392, "item_id": 50233587, "url": "https://news.ycombinator.com/item?id=50233587", "by": "k141", "date": "2026-10-25", "version": null, "stance": "recommend", "quote": "check out PEXEL and DeepSeek-OCR.", "excerpt": "yeah, there are models for that already, some are interesting, some are not. check out PEXEL and DeepSeek-OCR. in theory this could cut costs pretty well for long contexts + there no tokenisation layer.", "context": "", "use_case": "image-of-text models", "switched_from": "", "story": "Dessa: Conversation Model trained only on speech"}, {"id": 788722016, "item_id": 49295126, "url": "https://news.ycombinator.com/item?id=49295126", "by": "jubilanti", "date": "2026-08-13", "version": null, "stance": "recommend", "quote": "Deepseek was the first to really jump on this", "excerpt": "It sounds crazy, but it's way cheaper and faster for multimodal models to read text on images than the equivalent text represented as tokens. Text tokens take a lot more bytes to represent than you would think. There is no OCR-like intermediate step, it goes straight into the multimodal context where it's all just linea algebra underneath. The cost is more lossiness than if you gave it the same tokens as text. Which ", "context": "", "use_case": "", "switched_from": "", "story": "How Compaction Works in Pi"}, {"id": 780482432, "item_id": 48780152, "url": "https://news.ycombinator.com/item?id=48780152", "by": "calebkaiser", "date": "2026-07-03", "version": null, "stance": "recommend", "quote": "What makes the DeepSeek-OCR and related results exciting to some researchers", "excerpt": "Lots of researchers have done just this! There's a really rich history of research + lots of contemporary work on different encoding/representation strategies. This might be interesting to you: https://sbert.net/ What makes the DeepSeek-OCR and related results exciting to some researchers is less about the fact that you could devise a tokenization scheme that has fewer tokens, and more about how well it works.", "context": "", "use_case": "", "switched_from": "", "story": "60% Fable cost cut by converting code to images and having the model OCR it"}, {"id": 780445568, "item_id": 48777848, "url": "https://news.ycombinator.com/item?id=48777848", "by": "legel", "date": "2026-07-03", "version": null, "stance": "recommend", "quote": "DeepSeek-OCR proved 10x+ compression from visual embedding of text", "excerpt": "You are wrong. Text tokens are high-dimensional vectors, not 8 bits per character. Every token has a deep embedding, e.g. 1024 float values per text token. DeepSeek-OCR proved 10x+ compression from visual embedding of text, which was a groundbreaking result. [1] Very cool to see OP's project hacking on this principle. It's still not lossless, as noted in the github, but is a promising research direction. [1] https://", "context": "", "use_case": "", "switched_from": "", "story": "60% Fable cost cut by converting code to images and having the model OCR it"}, {"id": 778307712, "item_id": 48644232, "url": "https://news.ycombinator.com/item?id=48644232", "by": "ramon156", "date": "2026-06-23", "version": null, "stance": "recommend", "quote": "I love that the entire goal is to push Deepseek OCR further.", "excerpt": "I love that the entire goal is to push Deepseek OCR further. The west can learn greatly from these companies", "context": "", "use_case": "OCR", "switched_from": "", "story": "Unlimited OCR: One-Shot Long-Horizon Parsing"}]}], "caveats": [{"label": "", "summary": "", "quotes": [{"id": 790737412, "item_id": 49421088, "url": "https://news.ycombinator.com/item?id=49421088", "by": "phenomen", "date": "2026-08-24", "version": null, "stance": "warn", "quote": "PaddleOCR, dots.ocr and DeepSeek OCR performed significantly worse", "excerpt": "I tested many open-source and hosted OCR models and Datalab Chandra was the most accurate. It can parse complex layouts, tables, handwriting, and formulas at a fraction of the cost of Claude/Gemini. Local: https://github.com/datalab-to/chandra Hosted: https://www.datalab.to Another decent option is GLM OCR. It's slightly less accurate but faster and cheaper. Local: https://github.com/zai-org/GLM-OCR Hosted: https://d", "context": "", "use_case": "", "switched_from": "", "story": "OCR It \u2013 pull text out of un-copyable documents for your LLM"}, {"id": 752555520, "item_id": 47034720, "url": "https://news.ycombinator.com/item?id=47034720", "by": "kbyatnal", "date": "2026-02-16", "version": null, "stance": "warn", "quote": "Deepseek OCR is no longer state of the art.", "excerpt": "Deepseek OCR is no longer state of the art. There are much better open source OCR models available now. ocrarena.ai maintains a leaderboard, and a number of other open source options like dots [1] or olmOCR [2] rank higher. [1] https://www.ocrarena.ai/compare/dots-ocr/deepseek-ocr [2] https://www.ocrarena.ai/compare/olmocr-2/deepseek-ocr", "context": "", "use_case": "", "switched_from": "", "story": "Rolling your own serverless OCR in 40 lines of code"}]}], "who": [], "tags": [], "use_cases": ["ocr", "image-of-text models"], "compared": [{"slug": "dots-ocr", "name": "dots.ocr", "co": 2, "switched_from": 0, "switched_to": 0}], "makers": [], "threads": [{"url": "https://news.ycombinator.com/item?id=48776464", "title": "60% Fable cost cut by converting code to images and having the model OCR it", "mentions": 2}, {"url": "https://news.ycombinator.com/item?id=50213925", "title": "Dessa: Conversation Model trained only on speech", "mentions": 1}, {"url": "https://news.ycombinator.com/item?id=49415852", "title": "OCR It \u2013 pull text out of un-copyable documents for your LLM", "mentions": 1}, {"url": "https://news.ycombinator.com/item?id=49289654", "title": "How Compaction Works in Pi", "mentions": 1}, {"url": "https://news.ycombinator.com/item?id=48643426", "title": "Unlimited OCR: One-Shot Long-Horizon Parsing", "mentions": 1}, {"url": "https://news.ycombinator.com/item?id=46988538", "title": "Rolling your own serverless OCR in 40 lines of code", "mentions": 1}], "sentiment_by_month": {"2025-11": {"recommend": 1}, "2025-12": {"neutral": 1}, "2026-01": {}, "2026-02": {"warn": 1}, "2026-03": {}, "2026-04": {}, "2026-05": {}, "2026-06": {"recommend": 1}, "2026-07": {"recommend": 2}, "2026-08": {"recommend": 1, "warn": 1}, "2026-09": {}, "2026-10": {"recommend": 1}}, "versions": [], "version_unnamed": 0, "family_ref": null, "runs_on": null, "lines": [], "quotes": [{"id": 803737392, "item_id": 50233587, "url": "https://news.ycombinator.com/item?id=50233587", "by": "k141", "date": "2026-10-25", "version": null, "stance": "recommend", "quote": "check out PEXEL and DeepSeek-OCR.", "excerpt": "yeah, there are models for that already, some are interesting, some are not. check out PEXEL and DeepSeek-OCR. in theory this could cut costs pretty well for long contexts + there no tokenisation layer.", "context": "", "use_case": "image-of-text models", "switched_from": "", "story": "Dessa: Conversation Model trained only on speech"}, {"id": 790737412, "item_id": 49421088, "url": "https://news.ycombinator.com/item?id=49421088", "by": "phenomen", "date": "2026-08-24", "version": null, "stance": "warn", "quote": "PaddleOCR, dots.ocr and DeepSeek OCR performed significantly worse", "excerpt": "I tested many open-source and hosted OCR models and Datalab Chandra was the most accurate. It can parse complex layouts, tables, handwriting, and formulas at a fraction of the cost of Claude/Gemini. Local: https://github.com/datalab-to/chandra Hosted: https://www.datalab.to Another decent option is GLM OCR. It's slightly less accurate but faster and cheaper. Local: https://github.com/zai-org/GLM-OCR Hosted: https://d", "context": "", "use_case": "", "switched_from": "", "story": "OCR It \u2013 pull text out of un-copyable documents for your LLM"}, {"id": 788722016, "item_id": 49295126, "url": "https://news.ycombinator.com/item?id=49295126", "by": "jubilanti", "date": "2026-08-13", "version": null, "stance": "recommend", "quote": "Deepseek was the first to really jump on this", "excerpt": "It sounds crazy, but it's way cheaper and faster for multimodal models to read text on images than the equivalent text represented as tokens. Text tokens take a lot more bytes to represent than you would think. There is no OCR-like intermediate step, it goes straight into the multimodal context where it's all just linea algebra underneath. The cost is more lossiness than if you gave it the same tokens as text. Which ", "context": "", "use_case": "", "switched_from": "", "story": "How Compaction Works in Pi"}, {"id": 780482432, "item_id": 48780152, "url": "https://news.ycombinator.com/item?id=48780152", "by": "calebkaiser", "date": "2026-07-03", "version": null, "stance": "recommend", "quote": "What makes the DeepSeek-OCR and related results exciting to some researchers", "excerpt": "Lots of researchers have done just this! There's a really rich history of research + lots of contemporary work on different encoding/representation strategies. This might be interesting to you: https://sbert.net/ What makes the DeepSeek-OCR and related results exciting to some researchers is less about the fact that you could devise a tokenization scheme that has fewer tokens, and more about how well it works.", "context": "", "use_case": "", "switched_from": "", "story": "60% Fable cost cut by converting code to images and having the model OCR it"}, {"id": 780445568, "item_id": 48777848, "url": "https://news.ycombinator.com/item?id=48777848", "by": "legel", "date": "2026-07-03", "version": null, "stance": "recommend", "quote": "DeepSeek-OCR proved 10x+ compression from visual embedding of text", "excerpt": "You are wrong. Text tokens are high-dimensional vectors, not 8 bits per character. Every token has a deep embedding, e.g. 1024 float values per text token. DeepSeek-OCR proved 10x+ compression from visual embedding of text, which was a groundbreaking result. [1] Very cool to see OP's project hacking on this principle. It's still not lossless, as noted in the github, but is a promising research direction. [1] https://", "context": "", "use_case": "", "switched_from": "", "story": "60% Fable cost cut by converting code to images and having the model OCR it"}, {"id": 778307712, "item_id": 48644232, "url": "https://news.ycombinator.com/item?id=48644232", "by": "ramon156", "date": "2026-06-23", "version": null, "stance": "recommend", "quote": "I love that the entire goal is to push Deepseek OCR further.", "excerpt": "I love that the entire goal is to push Deepseek OCR further. The west can learn greatly from these companies", "context": "", "use_case": "OCR", "switched_from": "", "story": "Unlimited OCR: One-Shot Long-Horizon Parsing"}, {"id": 752555520, "item_id": 47034720, "url": "https://news.ycombinator.com/item?id=47034720", "by": "kbyatnal", "date": "2026-02-16", "version": null, "stance": "warn", "quote": "Deepseek OCR is no longer state of the art.", "excerpt": "Deepseek OCR is no longer state of the art. There are much better open source OCR models available now. ocrarena.ai maintains a leaderboard, and a number of other open source options like dots [1] or olmOCR [2] rank higher. [1] https://www.ocrarena.ai/compare/dots-ocr/deepseek-ocr [2] https://www.ocrarena.ai/compare/olmocr-2/deepseek-ocr", "context": "", "use_case": "", "switched_from": "", "story": "Rolling your own serverless OCR in 40 lines of code"}, {"id": 741619585, "item_id": 46351224, "url": "https://news.ycombinator.com/item?id=46351224", "by": "ooshaun", "date": "2025-12-21", "version": null, "stance": "neutral", "quote": "Built on DeepSeek-OCR", "excerpt": "We built PDF Craft, an MIT-licensed PDF to EPUB/Markdown converter that handles the hard parts: scanned documents, LaTeX formulas, and complex layouts. Key features: Accurate OCR: Built on DeepSeek-OCR, works even with low-quality scans LaTeX formula support: Precise recognition of math equations and chemical formulas Smart layout preservation: Handles two-column formats and text-image layouts Local or cloud: Run fre", "context": "", "use_case": "OCR", "switched_from": "", "story": "PDF Craft \u2013 Open-Source PDF to eBook Converter Powered by DeepSeek-OCR"}, {"id": 732553984, "item_id": 45784624, "url": "https://news.ycombinator.com/item?id=45784624", "by": "finnborge", "date": "2025-11-01", "version": null, "stance": "recommend", "quote": "images can be profoundly more token-efficient encodings of information than even CSVs", "excerpt": "If you haven't already seen the DeepSeek OCR paper [1], images can be profoundly more token-efficient encodings of information than even CSVs! [1]: https://github.com/deepseek-ai/DeepSeek-OCR/blob/main/DeepSeek_OCR_paper.pdf", "context": "", "use_case": "", "switched_from": "", "story": "Show HN: Why write code if the LLM can just do the thing? (web app experiment)"}]}