DeepSeek-OCR
Open document OCR that compresses pages into vision tokens.
About
DeepSeek-OCR is an open-source document OCR system that takes a different approach: it renders text as images and compresses pages into a small set of vision tokens instead of processing raw text tokens. It pairs a 380M-parameter DeepEncoder with a 3B MoE decoder, reaching 97% decoding precision at under 10x compression, and a single A100 can process 200,000+ pages per day.
Weights and code are public, though it is a research exploration of optical context compression rather than a production OCR service.
Specs
Context window
Pricing
Free quota
API access
Weights
Released
Benchmarks
97% OCR precision at <10x compression; beats GOT-OCR2.0 and MinerU2.0 on OmniDocBench (DeepSeek-reported).
Provider-reported unless a source says otherwise. Benchmarks are marketing-adjacent by nature — treat them as directional.
Use cases
Trust score — 7/10
Scored 7/10 (verified tier, metrics-v1). Evidence: novel optical context compression idea; 200k+ pages/day throughput claim; weights and code public on GitHub and Hugging Face; research exploration rather than a production OCR service. License: Weights and code public; license not stated in sources reviewed — check the repo..
How the trust score is built
Full methodology →- Adoption · stars/downloads, log-scaled vs peers
- 6.0/10
- Maintenance · recency, releases, issue load
- 6.0/10
- Openness · license + docs
- 9.0/10
- Manual review · sanity read + red-flag check
- 7.0/10
- · novel optical context compression idea; 200k+ pages/day throughput claim
- · weights and code public on GitHub and Hugging Face
- · research exploration rather than a production OCR service