Codesota · Models · DeepSeek-OCRDeepSeek4 results · 2 benchmarks
Model card

DeepSeek-OCR.

DeepSeekopen-sourceVision-Language OCR Model

DeepSeek's OCR model for document understanding.

§ 02 · Benchmarks

Every benchmark DeepSeek-OCR has a recorded score for.

#BenchmarkArea · TaskMetricValueRankDateSource
01olmOCR-BenchComputer Vision · Document Parsingaccuracy75.7%#15/18source ↗
02olmOCR-BenchComputer Vision · Document Parsingpass-rate75.7%#15/21source ↗
03olmOCR-BenchComputer Vision · Document Parsingpass-rate75.4%#16/21source ↗
04OmniDocBenchComputer Vision · Document Parsingcomposite86.5%#19/34source ↗
Rank column shows this model’s position vs all other models scored on the same benchmark + metric (competitors after the slash). #1 in red means current SOTA. Sorted by rank, then newest result.
§ 03 · Strengths by area

Where DeepSeek-OCR actually performs.

Computer Vision
2
benchmarks
avg rank #16.3
§ 04 · Papers

1 paper with results for DeepSeek-OCR.

  1. 2025-10-21· 1 result

    DeepSeek-OCR: Contexts Optical Compression

§ 05 · Related models

Other DeepSeek models scored on Codesota.

DeepSeek-V4-Pro Max
4 results · 1 SOTA
DeepSeek R1
671B MoE params · 10 results
DeepSeek-V3
7 results
DeepSeek-V3.2
6 results
DeepSeek-Coder-V2-Instruct
Unknown params · 4 results
DeepSeek-V4-Flash Max
4 results
DeepSeek-V3.2-Speciale
3 results
DeepSeek V3.5
685B MoE params · 2 results
§ 06 · Sources & freshness

Where these numbers come from.

pwc-dump
1
result
alphaxiv-leaderboard
1
result
github-readme
1
result
omnidocbench-leaderboard
1
result
0 of 4 rows marked verified.