Skip to content

DeepSeek · OCR

DeepSeek OCR 2

DeepSeek 🇨🇳OCRDeepSeek familyOpen weights. One of its sources says the weights are publicly available.
From Eurouter's listing:
DeepSeek-OCR 2 is DeepSeek's open-weight 3B vision-language OCR model. Its DeepEncoder V2 encoder reorders visual tokens (visual causal flow) rather than raster-scanning, converting documents, screenshots, receipts and scene text into text or layout-aware Markdown, including tables, formulas and reading order.
Max context
8K
Listings
7
Providers
7

Evidence and sources

Sources: Novita AI data, PPIO data, Models.dev data, EURouter data, Hugging FacePrimary data from Novita AI from PPIO: name from Models.dev: weights availability from EURouter: description from Hugging Face: release date
Tags from its sources:open-sourcevisiondocument-analysismultilingualmoe

Advantages and trade-offs

Compared with 10 representative OCR models

Advantages

Trade-offs

We only make a claim when there's enough data on models of the same type. Updated 22 Nov, 04:34 UTC.

Input & output

as stated by the model's providers
Text
Images
Text

Capabilities

Supported operations and provider-stated features

Vision

Listings

 302aideepseek/deepseek-ocr-2Source: 302.ai data8K$0.031$0.031—
 Eurouterdeepseek-ocr-2Source: EURouter data8K——
 Novita AIdeepseek/deepseek-ocr-2Sources: LiteLLM data, Models.dev data, Novita AI data8K$0.03$0.03—
 Ppiodeepseek/deepseek-ocr-2Source: PPIO data8K—
 Privatemode AIdeepseek-ocr-2Source: Models.dev data8K$0.8897cache read $0.0924$1.4675—

2 listings without a price are hidden.

Benchmarks

Named effort levels run from lowest to highest. Select a result to see its source details.

OCR

OmniDocBenchSource: OmniDocBench data

Vision

Other

Data credits: Hugging Face official benchmark leaderboards; licence varies by benchmark · OmniDocBench (OpenDataLab), Apache-2.0 (licence)

Also known as: deepseek-ai/deepseek-ocr-2, deepseek/deepseek-ocr-2, novita/deepseek/deepseek-ocr-2

Like this one. Tap or hover any dotted value, label or icon to see where it comes from and how it is measured. open details when tapped or hovered. How this data works