DeepSeek icon

DeepSeek

DeepSeek API for V4.1 Flash and V4-Pro models, with separate web Search and page Fetch endpoints.

Usage-based API2026年11月に掲載AI APILarge Language ModelWeb Search

DeepSeekとは?

DeepSeek provides API access to the deepseek-flash model (served by DeepSeek-V4.1-Flash-1017) and deepseek-v4-pro (DeepSeek-V4-Pro-0813), plus separate Search and Fetch APIs for live web content. The API supports OpenAI-compatible and Anthropic-compatible formats. DeepSeek’s Search API returns structured web results and its Fetch API retrieves one page as Markdown. The consumer chat route could not be accessed during this review, so its current availability and pricing were not verified.
Create an API key in the DeepSeek Platform, choose deepseek-flash or deepseek-v4-pro, and call the Chat Completions API using the OpenAI-compatible base URL or Anthropic-compatible endpoint. The docs list a 1M-token context length, up to 384K maximum output, thinking and non-thinking modes, JSON output, function tool calls, and Responses API support. Vision is listed for Flash but not Pro. Search and Fetch are separate bearer-authenticated endpoints billed per successful request.

DeepSeekの主な機能と特長

Use deepseek-flash, served by DeepSeek-V4.1-Flash-1017, or deepseek-v4-pro, served by DeepSeek-V4-Pro-0813.
Call the API using OpenAI-compatible or Anthropic-compatible formats and SDKs.
The models list a 1M-token context length and maximum output up to 384K tokens, subject to request constraints.
Both models support thinking and non-thinking modes, JSON output, tool calls, Responses API, and Anthropic API format.
The docs list vision support for deepseek-flash; deepseek-v4-pro is listed without vision support.
Use Search API to retrieve structured live web search results from DeepSeek’s in-house search engine.
Use Fetch API to retrieve one web page and return its content as Markdown for agents or retrieval workflows.
Connect DeepSeek as a backend to supported agent and coding tools; DeepSeek Harness is listed as a developer preview.

使い方

  1. 1Create an API key in the DeepSeek Platform and configure it in your application using an environment variable or secrets manager.
  2. 2Use https://api.deepseek.com for the OpenAI-compatible API or https://api.deepseek.com/anthropic for the Anthropic-compatible API.
  3. 3Select deepseek-flash or deepseek-v4-pro in a chat completion request and send messages; enable streaming or thinking mode if needed.
  4. 4For live retrieval, call POST /v1/tools/search with a query, then call POST /v1/tools/fetch for a selected URL when page content is needed.
  5. 5Check token charges, request charges, model version, and applicable limits in the official documentation before scaling usage.

活用例とアイデア

  • Build chat, coding, analysis, or agent applications using the compatible chat APIs.
  • Use deepseek-flash for tasks that need vision input or a lower-cost model route, subject to the current pricing and feature table.
  • Use deepseek-v4-pro for requests that need the Pro model’s listed capabilities.
  • Give an agent or retrieval-augmented generation workflow live web search results and fetched page text.
  • Connect DeepSeek models to compatible agent frameworks or coding assistants as a model backend.

制限事項と注意点

  • API access requires an API key and is usage-based; model token charges and Search/Fetch request charges are separate.
  • Prices and model versions can change. Check the live pricing page before making cost estimates.
  • Vision is listed for deepseek-flash only; deepseek-v4-pro is listed without vision support.
  • Search and Fetch are separate APIs, not automatically part of every chat request; successful requests are billed per request.
  • Concurrency limits in the captured rate-limit page are account-level: 2,500 for Flash and 500 for V4-Pro; exceeding the limit can return HTTP 429.
  • Legacy model names deepseek-v4-flash and deepseek-v4-flash-vision-exp are documented as retired and routed to the current Flash model at Flash pricing.
  • The official DeepSeek website rendered a client-side exception and chat.deepseek.com returned a CloudFront 403 during this review; web chat behavior, free access, and current UI platforms were not verified.
  • No official privacy or data-retention page was successfully reviewed in this session; privacy terms should be checked in the account and current official policies before sending sensitive data.

DeepSeekに関するよくある質問

  • The captured API documentation lists deepseek-flash (served by DeepSeek-V4.1-Flash-1017) and deepseek-v4-pro (DeepSeek-V4-Pro-0813).

  • Create an API key and use either the OpenAI-compatible base URL https://api.deepseek.com or the Anthropic-compatible endpoint https://api.deepseek.com/anthropic.

  • The Models & Pricing page lists a 1M-token context length and up to 384K maximum output, subject to request limits.

  • The pricing documentation lists vision support for deepseek-flash, but not for deepseek-v4-pro.

  • Yes. The Search API returns structured web search results, and the Fetch API retrieves a single page as Markdown. They are separate API endpoints.

  • The captured price list shows $0.75 per 1,000 successful Search requests and $0.75 per 1,000 successful Fetch requests.

  • By input and output tokens, with different rates for cache hits, cache misses, peak and off-peak hours, and model choice. See the live Models & Pricing page for current rates.

  • DeepSeek documents compatibility with supported agent and coding tools and lists DeepSeek Harness in developer preview.

  • The official chat route returned a CloudFront 403, and the main website rendered a client-side exception. Current consumer chat availability was not verified.

DeepSeek interface

製品仕様

料金プランUsage-based API
対応プラットフォームAPI (OpenAI-compatible), API (Anthropic-compatible)
主なカテゴリAI API

公式リソースとリンク