DeepSeek opens its app APIs: search, files, OCR and voice for developers
Updated November 14, 2026. Feature details and tool prices below refer to DeepSeek's own API.
On November 13, DeepSeek announced that developers can use the building blocks behind its chat app: web search, multi-step reasoning search, URL reading, file input, basic content safety checks, and standalone OCR, speech recognition and speech synthesis APIs. The practical change is that a developer can combine conversation and tools while DeepSeek handles search and page-reading orchestration on its service.
DeepSeek V4.1 Flash 1017 on MandAPI: BRL pricing
MandAPI's public catalog, checked November 14, 2026, lists R$0.39 per million input tokens and R$0.99 per million output tokens for deepseek-v4.1-flash-1017. These are competitive published BRL rates for testing the model without converting every token charge from dollars.
DeepSeek's official API price sheet lists the same version under deepseek-flash at US$0.15 per million uncached input tokens and US$0.60 per million output tokens during off-peak hours. At an illustrative R$5.00 per US$1, those become R$0.75 and R$3.00. In that example, MandAPI's published rates are 48% lower for input and 67% lower for output. Actual FX, taxes, cache terms and tool charges can change the complete bill.

Which DeepSeek API features are now available?
The deepseek-flash model supports the chat features through Messages, Responses and Chat Completions. With thinking disabled, deepseek_web_search performs web or image searches and returns citations. With thinking enabled, the model can search and read links over several steps. deepseek_read_url reads links in a user's message. File input accepts PDFs, Office documents and images for text or layout parsing. safety_check provides a basic input and output review. See DeepSeek's chatbot guide.
Search, Fetch, OCR, ASR and TTS also have standalone APIs, so an existing application can add just the functions it needs. DeepSeek released an open-source demo client that can import exported DeepSeek Chat history. The demo illustrates the API; it is not a production-ready service.
What do search, OCR and voice cost?
The official pricing page, checked November 14, quotes US dollars. Search and Fetch each cost US$0.75 per 1,000 uses, including model-triggered uses. File parsing and OCR cost US$0.225 per 1,000 pages. Promotional voice rates are US$0.058 per audio hour for ASR and US$0.225 per 10,000 characters for TTS, both 50% below list prices. The page does not specify an end date for the voice promotion. Model tokens are billed separately.
| Official tool | Unit | Published price |
|---|---|---|
| Search or Fetch | 1,000 uses of each API | US$0.75 |
| File parsing or OCR | 1,000 pages | US$0.225 |
| ASR, promotional rate | 1 audio hour | US$0.058 |
| TTS, promotional rate | 10,000 characters | US$0.225 |
These are DeepSeek prices, not MandAPI prices. Budget for token usage and tool usage separately. The independent Search and Fetch billing guide says a request is billable only when it succeeds and returns a result.
How can developers start?
DeepSeek's chatbot guide gives a streaming example with model: "deepseek-flash", deepseek_web_search and deepseek_read_url. The service performs the tool calls, but the client must present the result blocks and citations correctly. The open-source demo shows one way to build that interface.
What can MandAPI do today?
MandAPI lists deepseek-v4.1-flash-1017 in its BRL price catalog, alongside other models in a single gateway. That makes model costs easier to compare in reais. This article does not claim that MandAPI routes DeepSeek's new Search, Fetch, OCR, ASR, TTS endpoints or its app-specific chatbot parameters. Use DeepSeek's own endpoints for those tools. For compatible model calls through MandAPI, see the OpenAI-compatible API guide.
A gateway listing for a model does not automatically include every auxiliary service behind the provider's app. Choose the endpoint for the actual feature and test the complete workflow.
Related guides and next steps
- DeepSeek V4.1 Flash 1017 pricing, benchmarks and API goes deeper on the model (Portuguese).
- DeepSeek Super-agent architecture and prototype explores an agent that can combine models and tools.
- MandAPI model catalog helps compare available alternatives.
- The OpenAI-compatible API guide and first-call checklist lead to a test request (Portuguese).
FAQ
Can I recreate the DeepSeek app with these APIs?
DeepSeek supplies key building blocks and an open demo. A production app still needs its own interface, account handling, storage, safety measures and testing.
Are Search and Fetch free?
No. Each costs US$0.75 per 1,000 uses on the published price sheet, including model-triggered uses.
When does the voice discount end?
The current price sheet shows a 50% ASR and TTS promotion without an end date on that page. Check the live price before buying.
Can I call OCR or TTS through MandAPI?
This article does not announce support for those endpoints. Use DeepSeek's official API documentation for them and MandAPI's catalog for models offered through the gateway.