DeepSeek-OCR
Hacker ApprovedRanked 4 of 11 OCR models by recommenders
Free open-weights OCR model praised for compressing text into fewer tokens and strong results in use. Critics say newer models now beat it.
Free open-weights OCR model praised for compressing text into fewer tokens and strong results in use. Critics say newer models now beat it.
16 people gave an opinion
- recommend it
- 12
- criticize it
- 4
- approval share
- 75%
Recommended in 9 of 13 months
- November 2025: recommended
- December 2025: not recommended
- January 2026: not recommended
- February 2026: recommended
- March 2026: recommended
- April 2026: recommended
- May 2026: recommended
- June 2026: recommended
- July 2026: recommended
- August 2026: recommended
- September 2026: not recommended
- October 2026: recommended
- November 2026: not recommended (month in progress)
Why it carries the badge
- 12 people recommend itNeeds 10 or more, met.
- 3 recommenders for every criticNeeds 2 to 1 or better, met.
- Recommended in 9 of 13 monthsNeeds 4 or more, met.
The same three rules apply to every product, and the badge can’t be bought. How the badge works
What people say about it
The reasons people give, with the number of comments that give each one. Open a reason to read what they said.
Why people recommend it
Each mark is one comment
Visual compression cuts token costs
“DeepSeek-OCR proved 10x+ compression from visual embedding of text, which was a groundbreaking result.”
Comment from 3 Jul 2026 in “60% Fable cost cut by converting code to images and having the model OCR it” “in theory this could cut costs pretty well for long contexts + there no tokenisation layer.”
Comment from 25 Oct 2026 in “Dessa: Conversation Model trained only on speech” “images can be profoundly more token-efficient encodings of information than even CSVs”
Comment from 1 Nov 2025 in “Show HN: Why write code if the LLM can just do the thing? (web app experiment)” Works well in practice
“vLLM gave me enormous throughput improvements on big documents via continuous batching”
Comment from 3 Apr 2026 in “Google releases Gemma 4 open models” “which btw is amazing from my experimentations.”
Comment from 16 Feb 2026 in “Rolling your own serverless OCR in 40 lines of code” “Turns out it works really well”
Comment from 3 Jul 2026 in “60% Fable cost cut by converting code to images and having the model OCR it” Steady stream of new ideas
“Every major release from them has brought something really new to the field”
Comment from 18 Jun 2026 in “DeepSeek Introduces Vision” “I love that the entire goal is to push Deepseek OCR further”
Comment from 23 Jun 2026 in “Unlimited OCR: One-Shot Long-Horizon Parsing” “for their valuable models and ideas”
Comment from 23 Jun 2026 in “Unlimited OCR: One-Shot Long-Horizon Parsing”
What people warn about
Each mark is one comment
Outperformed by newer models
“Deepseek OCR is no longer state of the art.”
Comment from 16 Feb 2026 in “Rolling your own serverless OCR in 40 lines of code” “performed significantly worse”
Comment from 24 Aug 2026 in “OCR It – pull text out of un-copyable documents for your LLM”
Other OCR models people recommend
The four most recommended other OCR models, with the same counts for each, so you can compare like for like. The tick on each bar marks the 2 to 1 line the badge needs.
| Product | Recommend | Criticize | Approval | Months | Owned | Badge |
|---|---|---|---|---|---|---|
| Mistral OCR | 69 | 40 | 63% | 12 | — | — |
| PaddleOCR | 19 | 4 | 83% | 8 | — | Approved |
| GLM-OCR | 16 | 4 | 80% | 7 | — | Approved |
| DeepSeek-OCRThis page | 12 | 4 | 75% | 9 | — | Approved |
| dots.ocr | 8 | 1 | 89% | 5 | — | — |
What people said
14 of the 19 comments that mention it, newest first. Each links to the original comment.
“in theory this could cut costs pretty well for long contexts + there no tokenisation layer.”
“performed significantly worse”
“Turns out it works really well”
“DeepSeek-OCR proved 10x+ compression from visual embedding of text, which was a groundbreaking result.”
“I love that the entire goal is to push Deepseek OCR further”
“for their valuable models and ideas”
“Every major release from them has brought something really new to the field”
“smaller open-weights models that are dedicated for OCR tasks”
“vLLM gave me enormous throughput improvements on big documents via continuous batching”
“which btw is amazing from my experimentations.”
“Deepseek OCR is no longer state of the art.”
“Both versions of DeepseekOCR have the same problem.”
“They never actually test that pipeline.”
“images can be profoundly more token-efficient encodings of information than even CSVs”
Counts come from 1,795,640 Hacker News comments, November 2025 to November 2026. How we count. Report a mistake.