DeepSeek

DeepSeek is an AI platform built around a conversational assistant and a developer API. On the web and in mobile apps you can chat with DeepSeek, switch on DeepThink reasoning, or search for current information. Developers reach the same models through an API that accepts OpenAI and Anthropic request formats, and a desktop Harness app is also available. The chat product has no published subscription price, while API access is billed by token.

Key Features

• DeepThink, a reasoning mode for problems that benefit from step-by-step work.
• Search, which brings current web results into the conversation.
• Two API models, deepseek-flash and deepseek-v4-pro, with the newer Flash build tuned for lower cost and faster completion.
• A 1M-token context window and up to 384K output tokens on the API.
• Thinking and non-thinking modes, with thinking enabled by default.
• OpenAI-format and Anthropic-format base addresses, so existing SDK integrations can point at DeepSeek without a rewrite.
• JSON output, tool calls, the Responses API, Chat Prefix Completion, FIM completion, and Search and Fetch endpoints.

Who is it for

• Individuals who want a capable chat assistant without a subscription.
• Developers who need a low-cost API for chat, reasoning, or coding features.
• Teams moving an existing OpenAI or Anthropic SDK integration to another provider.
• Applications that work with long inputs, such as large documents or codebases.
• Builders who want web search and page fetching alongside the model.

What can you do with DeepSeek

• Ask a question in the chat and let DeepThink work through the reasoning before it answers.
• Run a search from the assistant when the answer depends on recent information.
• Call the API for chat completions, tool calls, structured JSON output, or code completion.
• Send long documents or repository context up to the 1M-token context limit.
• Add web search and page fetching to an application through the API endpoints.
• Download the Harness desktop app if you prefer a dedicated desktop client.

How does DeepSeek work

• Open the web chat or install the mobile app and sign in.
• Type a prompt, then turn on DeepThink or Search when the question needs reasoning or current sources.
• Read the answer and continue the conversation with follow-up questions.
• For development, create an API key and send requests to the OpenAI or Anthropic compatible base address.
• Choose deepseek-flash for speed and cost or deepseek-v4-pro for heavier work; spend is tracked by token usage.

Pricing

• The chat product on the web and in the mobile apps has no published subscription price.
• API usage is billed per million tokens. For deepseek-flash, cache-miss input is $0.15 off-peak and $0.30 at peak, and output is $0.60 off-peak and $1.20 at peak.
• For deepseek-v4-pro, cache-miss input is $0.66 off-peak and $1.32 at peak, and output is $1.98 off-peak and $3.96 at peak.
• Cached input costs far less: $0.003 per million tokens for deepseek-flash and $0.022 for deepseek-v4-pro off-peak.
• Search and Fetch calls cost $0.75 per 1,000 successful requests each.
• Off-peak rates apply outside weekday peak windows and are half the peak rates.

FAQ

Is DeepSeek free?

The chat product does not list a subscription price. API usage is paid, with prices listed per million input and output tokens.

What is DeepThink?

DeepThink is the reasoning mode. It works through a problem step by step before producing the final answer, which helps on multi-step tasks.

Which models does the API offer?

The API offers deepseek-flash, based on DeepSeek V4.1-Flash, and deepseek-v4-pro, based on DeepSeek V4 Pro. Both support thinking and non-thinking modes.

Can I use my existing OpenAI code?

Yes. The API accepts the OpenAI request format at one base address and the Anthropic format at another, so compatible SDKs can be repointed.

How large an input can I send?

The API supports a 1M-token context window and a maximum output of 384K tokens, with billing based on total input and output tokens.

Does the API include web search?

Yes. Standalone Search and Fetch endpoints are available and billed per 1,000 successful requests.