---
id: "b2"
language: "en"
library: "tools"
source_locale: "en"
source_market: "global"
official_url: "https://platform.deepseek.com/"
profile_updated_on: "2026-10-24"
---

# DeepSeek

A language-model service available through chat and compatible APIs for apps and agents.

- Company / organization: DeepSeek (Hangzhou)
- Company country / region: China · Company registration
- Category: Foundation models
- Official website: [https://platform.deepseek.com/](https://platform.deepseek.com/)
- Profile updated: 2026-10-24


## Overview

DeepSeek provides a chat interface and developer APIs for language-model questions, generation and reasoning-based applications. Developers send messages and instructions through OpenAI- or Anthropic-compatible formats, or configure it as the backend of coding assistants and agents. Its SDK examples and agent-integration guides suit teams adding model capabilities to existing software.

## Usage guide & costs

### Useful for

Use it as the model backend for apps, coding assistants or agents, with configurable thinking and tool calls. The foundation-model entry is for developers; the profile also links to chat.

[Official details](https://api-docs.deepseek.com/)
[Official details](https://api-docs.deepseek.com/guides/thinking_mode/)

### Access & setup

Obtain an API key, then configure the endpoint, model and messages. Current documented IDs are deepseek-flash and deepseek-v4-pro. Chat sign-in does not configure an API client.

[Official details](https://api-docs.deepseek.com/)

### Suggested workflow

Start with a small set of real tasks to compare thinking settings, latency and token use. Flash and Pro differ in vision support; check the model table before sending images.

[Official details](https://api-docs.deepseek.com/)
[Official details](https://api-docs.deepseek.com/guides/thinking_mode/)
[Official details](https://api-docs.deepseek.com/quick_start/pricing/)

### Billing model

API billing is based on input and output tokens; rates depend on the model and cache hits.

[Official details](https://api-docs.deepseek.com/quick_start/pricing/)

### Before you start

Search / Fetch has separate billing rules. The official page schedules per-request charges from November 1, 2026; it is currently free.

[Official details](https://api-docs.deepseek.com/quick_start/pricing/)


## Access options

### API platform · Foundation models

DeepSeek model APIs for reasoning, code, structured output and tool calls.

[API platform](https://platform.deepseek.com/)

### Web chat · AI assistants & clients

Chat directly in DeepSeek’s web app, using thinking for text and coding tasks.

[Web chat](https://chat.deepseek.com/)

## Capabilities

### Compatible API calls

Call models with messages and task instructions using documented OpenAI and Anthropic API formats and SDK configurations.

### Agent and coding integrations

Configure DeepSeek in compatible coding assistants or agent tools without building a separate chat interface.

## Official sources and references

- [Official website](https://platform.deepseek.com/)
- [Official website](https://chat.deepseek.com/)
- [Documentation](https://api-docs.deepseek.com/guides/thinking_mode/)

## Related product news

### DeepSeek updates V4.1-Flash-1017

- Original publication date: 2026-10-17

DeepSeek upgrades its web, app and API to V4.1-Flash-1017, retaining the deepseek-flash model ID and pricing. The update targets reasoning and agent tasks, with optimizations for DeepSeek Harness, Tencent CodeBuddy and WorkBuddy.

[DeepSeek](https://api-docs.deepseek.com/updates#deepseek-v41-flash-1017-update)

### DeepSeek opens its in-house search API

- Original publication date: 2026-10-17

DeepSeek opens Search / Fetch APIs for its own search engine, providing multilingual web search and page retrieval for models and agents.

[DeepSeek](https://api-docs.deepseek.com/updates#opening-up-the-in-house-search-api)

### DeepSeek releases V4.1-Flash

- Original publication date: 2026-09-10

DeepSeek releases V4.1-Flash with native visual understanding and an asymmetric MoE architecture aimed at reducing inference and cache costs. The multimodal API uses deepseek-flash; older Flash model IDs temporarily route to the new model.

[DeepSeek](https://www.deepseek.com/en/news/deepseek-v4-1-flash/)

## Reading & subscriptions

- [Interactive profile](../../../products/deepseek--b2.html)
- [Read](../../../read/en/products/deepseek--b2.html)
- [JSON](../../../read/en/products/deepseek--b2.json)

Descriptions summarize public information; consult official websites for features, pricing and regional availability.