Explore open models
Some DeepSeek models have openly released weights. R1 is released under MIT for research, running locally and further fine-tuning.
DeepSeek can be accessed through web chat, the mobile app, Harness and the API platform. Explore the available capabilities below and choose the entry point that fits your device or application.
DeepSeek brings conversation, multi-step reasoning, verifiable web search and coding into one assistant. Users can access the same capabilities in four ways.
The chat interface provides Deep Thinking and Smart Search switches. Reasoning is expanded by Deep Thinking; Smart Search brings checkable web content into replies.
Some DeepSeek models have openly released weights. R1 is released under MIT for research, running locally and further fine-tuning.
Harness is a desktop agent for documents, spreadsheets and code. Extensible plugins and background operation allow longer tasks to continue.
Your application can call the models through a request format compatible with mainstream SDKs. Both incremental streaming and structured output are available.
Drafting and revision sit alongside lengthy report summaries, comparisons and structured conclusions from raw material.
The model works through its reasoning before giving an answer. Mathematics, logical problems and tasks with several stages can use this approach.
Search combines answers with verifiable pages from the web. This provides current context for versions, news or prices rather than relying solely on internal knowledge.
All four access options use the same underlying models. Use context and installation requirements distinguish the options.
| Way to use | Where it runs | Does it need installation? | Where to start |
|---|---|---|---|
| Use DeepSeek Harness | Desktop devices | Requires the desktop application | Harness official download page |
| Use the DeepSeek API | Server-side and application use | No client required; register for a key | platform.deepseek.com |
| DeepSeek in your browser | Any browser-equipped device | None needed | chat.deepseek.com |
| DeepSeek App | iOS and Android mobile devices | Official-store installation needed | Look for the official app through App Store or major Android stores |
DeepSeek is a general-purpose AI assistant and model family developed by Hangzhou DeepSeek. The name may also appear as Deep Seek in English, without referring to a different product.
The web chat can be used free of charge. For API access, input/output token usage and the selected model determine charges, as listed on the platform.
The web version needs no download: open the official chat address in a browser. The official app is listed in App Store and mainstream Android stores; desktop users can visit the official Harness page.
Yes. Smart Search uses verifiable web sources in the browser app. Search and crawling are also exposed by the API; consult its official documentation for details.
Model selection, API integration and hands-on workflows are covered in the Chinese section.
Review how V4.1, V4 and R1 differ before choosing for your task.
Compare model capabilitiesFind executable curl/Python/JavaScript samples and common error checks.
Read API access instructionsThe six tutorials include prompt design, programming, document work, local deployment and API engineering.
Visit the tutorial directory