# deepseek-ai/DeepSeek-OCR

Contexts Optical Compression

Repository: https://github.com/deepseek-ai/DeepSeek-OCR
Canonical: https://ross.abutalabs.com/products/deepseek-ocr
Language: Python
License: MIT
License Family: permissive
Last push: 2026-01-27T03:45:14+00:00

## Health v2 (maintenance only)
Score: 45/100 (v2, computed 2026-09-02T17:46:02.011165+00:00)
- activity 64, release rhythm 35, longevity 22
- inputs: {"age_days": 320, "days_push": 218, "days_rel": null, "gap_med": null, "n_releases_24m": 0}
- flags: no_releases
- formula: round(0.45*activity + 0.35*rhythm + 0.20*longevity); archived -> min(score, 10)

## Adoption (not part of the score)
Stars 23855, forks 2203 (observed 2026-08-28T04:11:35.420777+00:00)

## What it is
DeepSeek-OCR is an open vision-language model from DeepSeek AI that researches 'contexts optical compression' - encoding long text contexts as images through a vision encoder feeding an LLM decoder. The repository ships model weights, inference code for both Transformers and vLLM, and an accompanying arXiv paper.

## Use cases
- extract text from scanned documents or images with an open-source OCR model
- compress long text contexts into images to extend LLM effective context
- run document understanding with a vision-language model
- serve an OCR model at scale with vLLM
- experiment with optical context compression research from the paper
- parse layouts, tables, and text from PDFs or screenshots

## When to choose
- you need high-accuracy open-weights OCR or document parsing with GPU acceleration
- you want to reproduce or build on the contexts-optical-compression research
- you want a DeepSeek vision model with official upstream vLLM support

## When to avoid
- you need a lightweight CPU-only OCR utility - this requires CUDA GPUs, flash-attention, and a heavy PyTorch/vLLM stack
- you want the newest iteration - DeepSeek-OCR2 was announced in January 2026 as the successor
- you need a turnkey OCR SaaS with preprocessing, batching UI, and export pipelines

## Facets
- artifact type: library
- maturity: active
- function: ocr, machine-learning, llm-inference, image-processing, computer-vision, transformers
- domain: artificial-intelligence, machine-learning, deep-learning, large-language-models, computer-vision, image-processing
- platform: python
- tags: vision-language-model, vision-encoder, optical-context-compression, context-compression, model-weights, inference-code, vllm, huggingface, multimodal, research-model, deepseek, mit-license, natural-language-processing, gpu, linux

## Member repositories
- deepseek-ai/DeepSeek-OCR (main) score 45

## Provenance
- Observed fields: from GitHub, fetched 2026-08-28T04:11:35.420777+00:00.
- Health v2: computed from the inputs above; adoption is never an input.
- Inferred fields (summary, facets, guidance): AI-extracted, prompt v1, taxonomy v1, on 2026-08-29T16:56:19.476607+00:00, confidence not recorded.
  - readme: https://github.com/deepseek-ai/DeepSeek-OCR (fetched 2026-08-28T04:11:35.420777+00:00, sha 4ce77cdea615)
- Data as of 2026-08-30T08:39:29.467469+00:00.