Back to Models
DeepSeekOpen Source

DeepSeek-OCR

DeepSeek-OCR is a visual large model developed by DeepSeek-AI. It has approximately 30.0B parameter scale and is released under the MIT license.

Parameters

30.0B

Context Window

License

MIT

Release Date

2025-10-20

API Pricing

API pricing for this model is not yet available

Strengths

  • 30-billion parameter scale
  • Available under an open license
  • Advanced visual information processing

Weaknesses

  • Specialization for specific tasks
  • Requires computational resources for operation
  • Lack of general-purpose dialogue capabilities

Use Cases

  • Optical character recognition from images
  • Visual data analysis
  • Document digitization

Deep Analysis

OCR Precision

97% at <10x compression

Vision Tokens

64-1853 per page

Production Speed

200k+ pages/day (single A100)

Languages

~100

License

Apache 2.0

Release Date

October 20, 2025

Strengths

  • Revolutionary compression (97% at 10x)
  • 200k+ pages/day on single GPU
  • ~100 language support
  • Deep parsing (charts, formulas)

Weaknesses

  • Not a general VLM
  • Degrades at 20x compression
  • No SFT stage (not a chatbot)

Competitor Comparison

Model
GOT-OCR2.0
MinerU2.0

DeepSeek-OCR pioneers optical compression: 97% precision at 10x compression. 200k+ pages/day on single A100, ~100 languages.

Sources

Analysis generated: 2026-05-24