DeepSeek-OCR 2 是 DeepSeek AI 发布的多模态文档识别模型,作为 DeepSeek-OCR 的升级版本,通过引入 DeepEncoder V2 架构,实现了视觉编码从「固定扫描」向「语义推理」的范式转变。该模型将原有 CLIP 编码器替换为轻量级语言模型(Qwen2-0.5B),并引入因果流查询机制,解码器沿用 DeepSeek-3B-MoE。模型仅需 256 至 1120 个视觉 Token 即可覆盖复杂文档页面,在 OmniDocBench v1.5 评测中综合得分达 91.09%,较前代提升 3.73%,阅读顺序识别编辑距离从 0.085 降至 0.057。
Model price
| Input /Mt | Output /Mt |
|---|---|
| Uncached | |
| ¥0.216/Mt | ¥0.216/Mt |
Sample Code
1base_url="https://api.ppio.com/openai"
2api_key="<YOUR_API_KEY>"
3
4curl "$base_url/v1/chat/completions" \
5 -H "Content-Type: application/json" \
6 -H "Authorization: Bearer $api_key" \
7 -d @- << 'EOF'
8{
9 "model": "deepseek/deepseek-ocr-2",
10 "messages": [
11
12 {
13 "role": "user",
14 "content": "Hi there!"
15 }
16 ],
17 "response_format": { "type": "text" }
18}
19EOF
20 Model Information
Context window:
8.2K
Maximum output tokens:
8.2K
Modality:
Supported features
API Integration Guide
Base URL:
https://api.ppio.com/openai
Messages protocol Base URL:
https://api.ppio.com/anthropic
API Endpoints:
/v1/chat/completions
Model ID:
deepseek/deepseek-ocr-2
API Key:
Create & retrieve your API KeyRate Limits
| Usage tier | RPM | TPM |
|---|---|---|
| T1 | 5 | 50000000 |
| T2 | 100 | 50000000 |
| T3 | 1000 | 50000000 |
| T4 | 3000 | 50000000 |
| T5 | 6000 | 50000000 |
Common tool integrations
Feature tutorials