AI News · 4 min read
DeepSeek puts search, files and voice on its API
On November 13, 2026, deepseek-flash gained the app's search, file, URL, safety, and voice tools. Tool fees sit on top of token charges.
What DeepSeek opened on November 13
On November 13, 2026, DeepSeek published The full DeepSeek App experience, now via API ↗. V4.1-Flash, model id deepseek-flash, now runs the app's tools inside Messages, Responses, and Chat Completions. Tool calls and context assembly happen on DeepSeek's servers.
The picture above is ours.
| Tool | What the announcement says |
|---|---|
| Web Search | Search the web or images, with citations, in non-thinking mode |
| Deep Thinking Search | Search and read pages through multi-step reasoning |
| File Input | Parse PDFs, Office files, images, and more into the conversation |
| URL Reading | Fetch links that appear in the user message |
| Safety Checks | Screen inputs and outputs, and return a refusal indicator when a risk is detected |
Turn search on with deepseek_web_search. Turn URL reading on with deepseek_read_url. Public-facing apps must enable safety checks. DeepSeek says the developer remains responsible for content moderation, applicable model registration, and labeling of AI-generated content.
Search, Fetch, OCR, ASR, and TTS are also standalone APIs. An open-source front end can import DeepSeek conversation history. The running demo is cdn.deepseek.com/chat-demo ↗, the code is deepseek-ai/chat-demo ↗, and the guide is Build a chatbot ↗.
What the tool calls cost
Tool fees sit on top of token charges. These are the figures on the pricing page ↗ as read on November 13. DeepSeek reserves the right to adjust them.
| Tool | Unit | Price on the pricing page |
|---|---|---|
| Search (text or image) | 1,000 searches | $0.75 |
| Web fetch | 1,000 fetches | $0.75 |
| File parsing and the OCR API | 1,000 pages | $0.225 |
| Text to speech | 10,000 characters | $0.45 list, $0.225 on the limited-time 50% discount |
| Speech to text | 1 hour of audio | $0.116 list, $0.058 on the limited-time 50% discount |
Search and web fetch bill per use, on a direct API call or when the model calls the tool. The speech-to-text guide ↗ bills to the second and rounds a fraction up to one second. The text-to-speech guide ↗ counts a Chinese character as two billed characters, and a letter, digit, or space as one.
File parsing bills once per successful page at upload. Later chats bill input tokens and do not bill the parse again. A failed upload is free. Plain text, spreadsheets, and ebooks sent as user_data are free. The file guide ↗ caps a request at 20 documents and 8 MiB of text, and you cannot download the parsed file.
Speech to text takes 16-bit PCM WAV or Ogg Opus, at most 8 MiB and 120 seconds, with no streaming. Text to speech is a full-text POST or a WebSocket. The voices are stella (default) and tide.
Token prices for deepseek-flash are a separate peak and off-peak table on that pricing page. Vercel AI Gateway's Ultrafast route is another host, at $2.40 in and $9.60 out per million tokens. That price is not the tool card above.
Who should call it
Build on it if the product is a chatbot or a retrieval app on deepseek-flash and you have been bolting on search, files, and voice. A public app still has to enable safety checks. Model registration and AI-content labeling stay with the developer.
DeepSeek Harness is the agent product and the plugin plan. The longer buyer note is the DeepSeek review. The weight comparison is best open-source LLMs 2026.
What people are saying
"Anyone can clone the DeepSeek app now. The app was never the moat." — Robert Yang (@mrrobertyang) ↗
"The part worth reading is the tool fee. Web search is $0.75 per 1,000 calls, on top of tokens." ... "Search-heavy agents will feel it." — Tengkolok (@Tengkolok24) ↗. The $0.75 figure matches the pricing page.
"that glue is what eats most of the time when you build a small AI app" — AI Adventurer (@AiAdventurerx) ↗
"Curious if the open-source demo's TTS is good enough to voice game characters." — IA en Crudo (@IAenCrudo) ↗
Reddit was not reachable for this story.
Our take
Call deepseek-flash with the built-in tools if the product is a chatbot that needs search, a document, or a voice, and you already send that traffic to DeepSeek. Set deepseek_web_search and deepseek_read_url. Enable safety checks before the app is public. AI Adventurer's line is the practical point: the time went into glue, and the announcement is DeepSeek offering that glue on the server.
Read the tool fee before the token price. Tengkolok has the line that matters. A thousand searches are $0.75 plus the tokens those pages add. File parsing is $0.225 per thousand pages, once, when the upload succeeds. The speech discounts end at the list prices in the table, $0.45 and $0.116. DeepSeek reserves the right to change the card.
Treat the demo as a shell, and listen before you voice a game. It imports conversation history. Text to speech has Stella and Tide. We have not graded them. IA en Crudo's question is a listening test, not a spec on the pricing page.
Keep Harness and Ultrafast separate. Harness is the agent product with the plugin plan. Ultrafast is Vercel's speed tier at the price in that post. This page is the official tool card.