DeepSeek opens the full App experience via API: built-in search, files and URL reading on V4.1-Flash, plus standalone OCR, ASR and TTS with separate tool fees
- Published
- 2026-11-13
- Type
- Product
- Organization
- DeepSeek
- Source
- DeepSeek
Why it matters
【Key points】
· Release: On 13 Nov 2026, DeepSeek’s API changelog opened the App/web stack to developers; @deepseek_ai posted it at 14:24 UTC. Not a new model; it targets existing deepseek-flash (V4.1-Flash-1017).
· In-chat tools: On Messages, Responses and Chat Completions, the service can run web or image search with citations, multi-step search and link reading, parsed file uploads, URL fetching, and an optional safety check that returns a refusal marker on risk.
· Standalone APIs: Search, Fetch, OCR, ASR and TTS can also be called alone. An official open-source demo can import chat history.
- · Pricing: Tool use is billed on top of tokens.
- Search and web fetch are $0.75 per 1,000, same as the 17 Oct list price, now also when used as tools.
- File parsing / OCR is $0.225 per 1,000 pages.
- TTS is $0.45 per 10,000 characters ($0.225 at limited-time 50% off); ASR is $0.116 per audio hour ($0.058 at 50% off).
- No discount end date is given.
【Our read】
Background
- These capabilities already formed the consumer DeepSeek experience.
- On 17 Oct the lab opened only its in-house search engine as Search and Fetch APIs.
- This drops that product stack into the existing API.
Comparison
This is closer to the built-in web, file and voice tools already offered by OpenAI and Anthropic, without a new model. Search and fetch prices are unchanged; new costs are file/OCR, speech, and how often tools fire.
Impact
- Teams can drop a layer of external search, OCR and speech.
- Tool fees stack on already-low Flash token prices, and deep-thinking search can multiply the bill.
- For public-facing apps, the safety parameter and China registration/labeling duties are part of the design.
- This is not a model leap.
【Verifiable】
Related developments
AGI Pulse reviews each article before publishing. Monitoring and drafting are AI-assisted. Each article adds original analysis: background, comparison, and impact under Our judgment, plus a source list you can check. Full source text is not republished. Editorial policy