本文来源:http://himiao.ai/en/news/2026-11-15-deepseek-full-app-api
若有版权问题,请来信告知:13951431913#139.com 或在右方提交评论。
DeepSeek 于 11 月 13 日宣布,把其 App 背后的整套能力——网页搜索、文件输入、链接读取、语音与安全检查——通过 API 开放给开发者,底层模型为 deepseek-flash(DeepSeek-V4.1-Flash-1017)。
发布分两个层面:一是内置参数,开发者在 Messages、Responses 与 Chat Completions 三类接口中开启相应开关,模型调用即可附带联网搜索、文件解析、链接读取与安全检查,工具调用与上下文拼装由服务端完成;二是独立接口,搜索与抓取、OCR、语音识别与语音合成都可以单独调用并按需组合成流水线。
搜索分两种模式:关闭思考时,模型按问题需要执行网页或图片搜索并给出带引用的回答;开启思考时,会跨多个推理步骤自行搜索并读取链接。文件输入支持 PDF、Office 文档与图片;语音识别支持 21 种语言自动检测,语音合成提供 10 种语言的两种音色;OCR 单文件上限 100 MB、最多 1000 页。DeepSeek 还开源了一个可导入 App 历史记录的演示聊天客户端。
定价方面,联网搜索约每千次成功调用 0.75 美元,OCR 每千页 0.225 美元,语音识别限时每小时 0.058 美元,语音合成每万字符 0.225 美元,均叠加在 token 费用之上。模型本身高峰期每百万输入 token 0.30 美元、输出 1.20 美元,非高峰减半,上下文窗口最高 100 万 token。这些内置能力目前仅随 deepseek-flash 提供,尚不支持 deepseek-v4-pro。
本文来源:http://himiao.ai/en/news/2026-11-15-deepseek-full-app-api
若有版权问题,请来信告知:13951431913#139.com 或在右方提交评论。
09:00 Nov 15 Upgraded to a story: expanded in place from this flash with API structure, pricing and background
Upgraded to a story: expanded in place from this flash with API structure, pricing and background
The release works on two levels. First, built-in parameters: developers switch on the relevant options in the Messages, Responses and Chat Completions APIs, and web search, file input, link reading and safety checks come along with the model call, with tool calls and context assembly handled server-side. Second, standalone APIs: search and fetch, OCR, speech-to-text and text-to-speech can be called separately and combined as a pipeline requires.
Search comes in two modes. With thinking off, the model runs web or image searches as the question requires and answers with citations; with thinking on, it searches the web and reads links on its own across multiple reasoning steps. File input accepts PDFs, Office files and images, adding them to the conversation context after plain-text extraction or rich-layout parsing. Link reading runs through the deepseek_read_url tool: the model spots links in the user message and fetches the pages itself.
Speech and text recognition are offered as standalone APIs. Speech-to-text supports 21 languages with automatic detection, takes clips of no more than 120 seconds and does not stream; text-to-speech offers two voices in 10 languages and can stream over WebSocket. The OCR API handles images and PDF, Word and PowerPoint documents, with a 100 MB maximum per file and up to 1,000 pages, returning results page by page; files processed through it are not stored in the Files API or passed to the chat model.
DeepSeek also open-sourced a demo chat client built on these APIs, which can import conversation history exported from the DeepSeek app. The built-in features ship only with the deepseek-flash model — DeepSeek-V4.1-Flash-1017 — and do not yet support deepseek-v4-pro; the model offers a context window of up to one million tokens, a maximum output of 384,000 tokens, and vision input.
In its Nov 13 changelog, DeepSeek said that web search, file input, URL reading, voice interaction and safety checks work together with model inference to form the conversation experience users know, adding: "Today we open these capabilities to developers." The docs also note that public-facing services must enable the safety check and meet obligations such as content safety review, large-model registration and AI labelling.
On pricing, WebProNews, citing DeepSeek's pricing pages, reported web search at $0.75 per 1,000 successful requests and OCR at $0.225 per 1,000 pages; speech-to-text at $0.058 per hour of audio for a limited time (list $0.116) and text-to-speech at $0.225 per 10,000 characters. The deepseek-flash model itself runs $0.30 per million input tokens and $1.20 per million output tokens at peak, half that off-peak; tool fees are charged on top of token costs.
The release builds on earlier steps. On Oct 17, 2026, DeepSeek opened its in-house search engine as a Search and Fetch API, saying it handles hundreds of millions of search requests from around the world every day and indexes hundreds of billions of web pages; the same day, V4.1 Flash was upgraded to the V4.1-Flash-1017 snapshot, scoring 42.6 on HLE (without tools), 42.3 on Terminal-Bench 4.0 and 77.0 on DeepSWE v1.1.
For comparison, OpenAI's Responses API and Anthropic's Messages API also offer web-search tools. DeepSeek has bundled search, files, voice and safety checks into one release, with tools billed per use or per volume separately from tokens, and developers choosing between built-in parameters and standalone endpoints.
Release: Nov 13, 2026, in the DeepSeek API docs changelog, 'Releasing the full suite of application APIs'
Model: deepseek-flash (DeepSeek-V4.1-Flash-1017); the built-in features do not yet support deepseek-v4-pro
Context: deepseek-flash offers a context window of up to one million tokens, a maximum output of 384,000 tokens, and vision input
Endpoints: Built-in features work across the Messages, Responses and Chat Completions APIs
Standalone APIs: Search and fetch, OCR, speech-to-text (ASR) and text-to-speech (TTS) can be called separately
OCR pricing: $0.225 per 1,000 pages; 100 MB maximum per file, up to 1,000 pages
Voice pricing: Speech-to-text $0.058 per hour of audio for a limited time (list $0.116); text-to-speech $0.225 per 10,000 characters
Model pricing: deepseek-flash at peak: $0.30 per million input tokens and $1.20 per million output tokens; half that off-peak
Demo: An open-source chat demo client that can import DeepSeek chat history
Background: On Oct 17, 2026, the in-house Search and Fetch API opened and V4.1 Flash was upgraded to V4.1-Flash-1017
Sources DeepSeek API Docs (Nov 13, 2026); WebProNews (Nov 14, 2026); Crypto Briefing (Nov 13, 2026); AI News Way (Nov 13, 2026). This article is compiled from public information and does not constitute investment advice.
A ping on your phone or computer when the brief is out. No email needed.
Oct 18 DeepSeek Releases V4.1-Flash-1017: MIT-Licensed Weights, In-House Search API
Oct 22 DeepSeek Open-Sources DeepInfer, the Inference Engine Behind Its Production Services
Oct 19 dsh Free-Model Plugin Hits No. 2 on GitHub Daily Trending: Frontier Models Without Login or API Key
Oct 8 One Man, One AI Stack, Nine Banks: CrowdStrike Details the AI-Driven Korean Bank Hacks
Oct 6 DeepSeek Nears $12B Funding Round with Tencent and CATL, Doubling Its Original Target
Brief Malaysia Probes Megaspeed Chip Re-Exports as Singapore Files 41 Charges 10 stories Listen
GitHub GitHub Weekly Vol.5: 6 Projects Worth a Look This Week Tested or read — weekly All →
Quiz Three questions on today’s news — then see where you rank 30 seconds Play →
Call Will Inkling-2-Small pass 280 likes on Hugging Face by Nov 16, 12:00 Beijing time? The cat says “Yes” Predict →
Nov 15 SemiAnalysis Names Nvidia's Seven Biggest GPU Customers; Six Build Their Own Chips
Nov 15 Malaysia Probes Megaspeed Chip Re-Exports as Singapore Files 41 Charges
Nov 15 Perplexity Open-Sources Decider 1.3, Adding Text-Span Extraction
Nov 15 Microsoft 365 Outage Lasts Over 24 Hours After Network Upgrade Fault
Nov 15 GitHub Copilot to Retire Six Models on Dec 14, With No Successor Yet for Kimi K3
Nov 15 Fed Report: AI Becomes the Top Financial-Stability Risk for the First Time
