Vision
Analyze supplied images together with written instructions.
All API models
DeepSeek/ LLMPlanned
DeepSeek/ LLMPlannedDeepSeek OCR API
deepseek-ocrDeepSeek OCR turns one document image into editable text or Markdown. Upload a scanned page, receipt, table, or figure and choose a focused extraction prompt. Each request processes one image independently, with up to 8K input and 8K output tokens.
- Vision
- Document OCR
- Streaming
This route is published for integration planning and is not accepting requests yet.
Playground
Configure a request, inspect the exact JSON, then run it with your account.
Playground opens when this route becomes available.
Response
The API response will appear here.
API PRICING
API pricing
Compare the complete API billing table for this route. Token models separate input, output, cache-read, cache-write, and long-context rates; media models vary by configuration.
Pricing coming soon
CAPABILITIES
What this route supports
Use one stable PixMind model ID while keeping provider-specific setup out of your application.
Document OCR
Use Document OCR through the live DeepSeek OCR route without maintaining a separate provider integration.
Streaming
Return partial output as it is generated for responsive product experiences.
WHY DeepSeek OCR
Production-grade image capabilities
What makes DeepSeek OCR a strong choice for high-fidelity, text-aware image workflows.
Document image to Markdown
Upload one scanned page and use Convert the document to markdown. to extract editable content.
Receipts, tables, and figures
Use Free OCR. for plain text or Parse the figure. for a table or chart image.
Simple token pricing
Studio and API both charge 21 input credits and 84 output credits per million tokens, at 50% of official rates.
TECHNICAL SPECS
Supported parameters
Inputs, supported configurations, and request parameters for the DeepSeek OCR API.
- Input
- One image, one user message
- Maximum image size
- 10 MB
- Context window
- 8,192 tokens
- Maximum output
- 8,192 tokens
- Output
- Editable text or Markdown
MODEL WORKFLOWS
What to build with DeepSeek OCR
Choose a task that matches the live route capabilities, then adapt one of these prompt starters.
Best-fit tasks
- Build drafting, rewriting, extraction, classification, and conversational workflows.
- Analyze supplied images together with written instructions.
- Return partial output as it is generated for responsive product experiences.
Prompt starters
Replace the bracketsStarter
Inspect the supplied image for [task]. Separate direct observations from inferences, extract [required fields], and return the result as [format].
Starter
Answer [user task] in a clear structure. Put the direct result first, keep each section self-contained, and avoid delaying essential information until the end.
MODEL SELECTION
Choose DeepSeek OCR when the job fits
Compare nearby live routes by task, supported capability, and current entry price before you lock a model ID into production.
| Model | Best for | Key capabilities | Entry price | Availability | Action |
|---|---|---|---|---|---|
DeepSeek OCR DeepSeek · current page | Build drafting, rewriting, extraction, classification, and conversational workflows. | Text · Vision · Image input | $0.02222 / 1M tokens | Planned | Selected |
DeepSeek V4 Flash DeepSeek | Build drafting, rewriting, extraction, classification, and conversational workflows. | Text · Reasoning · Coding | $0.237 / 1M tokens | Available | View route |
DeepSeek V4 Pro DeepSeek | Build drafting, rewriting, extraction, classification, and conversational workflows. | Text · Reasoning · Tool calling | $0.6667 / 1M tokens | Available | View route |
DeepSeek V4.1 Flash DeepSeek | Build drafting, rewriting, extraction, classification, and conversational workflows. | Text · Reasoning · Tool calling | $0.1481 / 1M tokens | Available | View route |
Capabilities and starting prices are a current PixMind route snapshot. The accepted request contains the final configuration and billing quote.
GET STARTED
From playground to production
Create a scoped key
Create an API key with the permissions required by this model.
Send the request
Use the model ID and endpoint shown above from your backend service.
Track the task
Read language-model output from the response. For image and video requests, use the returned task ID to retrieve the result.
FAQ
DeepSeek OCR API questions
What can DeepSeek OCR extract?+
Printed document content, receipts, table text, and figure descriptions from a single image.
Can I send a PDF or multiple pages?+
Send one image in one user message per request. Convert PDF pages to images and recognize each page separately.
Which prompt should I use?+
Use Free OCR. for text, Convert the document to markdown. for Markdown, and Parse the figure. for figure or table content.
What does it cost?+
Studio and API use the same all-day price: 21 credits per million input tokens and 84 credits per million output tokens, at 50% of official rates.
What model ID should I use for the DeepSeek OCR API?+
Use deepseek-ocr in the model field. This is the stable PixMind route identifier shown in the playground request JSON.
Which endpoint does the DeepSeek OCR API use?+
Send POST requests to /v1/chat/completions. Keep your API key on your server and pass it as a Bearer token.
Is the playground using a real API request?+
Yes. When you click Run model, PixMind forwards the displayed payload with the API key you entered. A successful request may create a paid task.
Does PixMind save the API key entered here?+
No. The playground keeps the key only in component memory and sends it for the current request; it is not written to browser storage.
Is API pricing the same as PixMind Studio pricing?+
No. API usage resolves through the apiPlatform pricing role and a separate pay-as-you-go ledger. Studio pricing includes the creation interface and product workflow services.
Why can the final price differ from the starting price?+
The entry price reflects the configuration displayed on this page. Resolution, duration, quality, audio, references, and output count can affect the final task price.
RELATED MODELS
Compare other API routes
DeepSeek
DeepSeek V4 Flash
Cost-efficient DeepSeek V4 text route for high-volume extraction, classification, rewriting, summaries, and lightweight coding with a one-million-token context.
View APIDeepSeek
DeepSeek V4 Pro
DeepSeek V4 Pro 0813 handles demanding code review, architecture decisions, and long-document reasoning. Give it the source material and acceptance criteria, then use tools to turn the plan into changes. The text route supports a 1M-token context and up to 384K output tokens.
View APIDeepSeek
DeepSeek V4.1 Flash
DeepSeek V4.1 Flash 1017 turns a detailed project brief into working code and supports iterative implementation with tools. Use it for prototypes, repository changes, and repeated document extraction when you need long context at a lower token price.
View APISTART WITH THE CONTROL PLANE
Turn your first DeepSeek OCR prompt into a real API request
Test the exact payload in the playground, then create a scoped key and move the same model ID into your application.