What Is DeepSeek?
DeepSeek is an AI company in Hangzhou, China, and the name of its AI assistant. Users can chat with DeepSeek on the web or in its mobile app, and developers can connect to its API. The models have a thinking mode for harder problems, and deepseek-flash can read images. DeepSeek also says it open-sources models such as DeepSeek-V4 and DeepSeek-R1.
Who Should Use DeepSeek?
DeepSeek is useful for people who want an AI chat assistant to try for free and for developers who want a low-cost API. The API works with the OpenAI and Anthropic formats, so many existing tools can connect with small changes. Teams should check the privacy terms first, since they were not reviewed here.
How Does DeepSeek Work?
- Open DeepSeek on the web or in the mobile app to chat. Developers can instead apply for an API key on the DeepSeek Platform.
- Type your question or task. Use the thinking or search options on the home page when the task needs them.
- Developers send requests to https://api.deepseek.com with the model name deepseek-flash, or to the Anthropic-format address.
- Read the answer, then continue the conversation or send follow-up requests.
DeepSeek Pricing and Plans
The screenshot shows DeepSeek's API pricing, not a monthly app plan. It lists two models, deepseek-flash and deepseek-v4-pro, with prices per 1M tokens. A token is a small piece of text, such as a word or a number. Off-peak prices are half of peak prices. Peak hours are 01:00 to 04:00 and 06:00 to 10:00 UTC, Monday through Friday, except Chinese public holidays. DeepSeek's September news post says deepseek-v4-pro requests are billed at the Flash price until V4.1 Pro launches.
| Plan | Price | Billing | Main Features and Limits |
|---|---|---|---|
| deepseek-flash (API) | $0.15 input (cache miss) and $0.6 output per 1M tokens, off-peak | Not stated (billed by tokens used; peak prices are $0.3 and $1.2) | Cache-hit input $0.003 off-peak, $0.006 peak; 1M context; 384K max output; vision; concurrency limit 2500 |
| deepseek-v4-pro (API) | $0.66 input (cache miss) and $1.98 output per 1M tokens, off-peak | Not stated (billed by tokens used; peak prices are $1.32 and $3.96) | Cache-hit input $0.022 off-peak, $0.044 peak; 1M context; 384K max output; no vision; concurrency limit 500 |
Is DeepSeek Free?
The official download page says the DeepSeek app lets users try the AI assistant for free. The pricing page does not list a price or limits for the web chat or app. The API is separate and charges by tokens used. The pricing page mentions a granted balance but does not say who gets one.
DeepSeek Key Features
Thinking and Non-Thinking Modes
Both models support thinking and non-thinking modes, and thinking is the default. Thinking mode lets the model reason before it answers, which can help with harder problems. FIM Completion, a beta tool that fills in missing code, works in non-thinking mode only.
Long Context and Output
Both models list a context length of 1M, which is how much text the model can read in one request. The maximum output is 384K. These limits suit long documents, large code files and long conversations, though longer inputs use more tokens and cost more.
Vision on deepseek-flash
The deepseek-flash model accepts images with text and can describe pictures, read text from screenshots and analyze charts. JPEG, PNG, GIF and WebP files are supported. DeepSeek lists vision as not supported on deepseek-v4-pro.
OpenAI and Anthropic API Formats
The API works with the OpenAI and Anthropic formats, so developers can change the base URL and keep their current tools. It lists JSON output, tool calls and the Responses API. DeepSeek says tools such as Claude Code and OpenCode can use it as a backend model.
Search, Fetch and Harness
DeepSeek offers standalone Search and Fetch APIs, each priced at $0.75 per 1,000 successful requests. It also offers DeepSeek Harness, a desktop app in developer preview. DeepSeek says it can run in the background and read and write local files.
Should You Use DeepSeek?
DeepSeek fits people who want to try a free AI assistant and developers who want low per-token prices with a 1M context window. The screenshot covers only API pricing, and DeepSeek says deepseek-v4-pro requests are billed at the Flash price for now. Check the privacy and data terms, which this overview did not review, before sending work files. This overview is based on publicly available product information and is not a hands-on review.
Pros and cons
Pros
- The official download page says users can try the DeepSeek app for free.
- API prices are listed per 1M tokens, and off-peak prices are half of peak prices.
- Both models list a 1M context length and a 384K maximum output.
- The API works with both the OpenAI and Anthropic formats.
Cons
- DeepSeek reserves the right to change prices, and the pricing page tells users to check it regularly.
- The deepseek-v4-pro model does not support vision, and DeepSeek says its requests are billed at the Flash price for now.
- Peak hours cost twice as much as off-peak hours, so timing can change the cost of API work.
- Chat Prefix Completion and FIM Completion are beta features, and FIM works in non-thinking mode only.
Popular DeepSeek use cases
Ask Questions and Research
Users can chat with DeepSeek on the web or in the app and use its thinking and search options.
Read Images and Screenshots
Developers can send images to deepseek-flash to describe pictures, read text from screenshots or analyze charts.
Build Apps With the API
Developers can call the API for chat, JSON output and tool calls, using the OpenAI or Anthropic format.
Power Coding Agent Tools
Developers can use DeepSeek as the backend model in coding tools such as Claude Code and OpenCode.
Frequently asked questions
Is DeepSeek free?
The official download page says users can try the DeepSeek app for free. The API is separate and billed by tokens used. The pricing page does not list a price for the web chat or app.
What are the limits of the DeepSeek Free plan?
DeepSeek does not list a Free plan with set limits on the pages checked. The screenshot shows API concurrency limits instead: 2500 for deepseek-flash and 500 for deepseek-v4-pro. Chat usage limits are not published there.
What is the difference between deepseek-flash and deepseek-v4-pro?
Both list a 1M context length, a 384K maximum output and thinking and non-thinking modes. Flash supports vision and has a concurrency limit of 2500. Pro does not support vision and has a limit of 500. Pro costs more per token.
What are peak and off-peak hours in DeepSeek's API pricing?
Peak hours are 01:00 to 04:00 and 06:00 to 10:00 UTC, Monday through Friday, except Chinese public holidays. All other hours are off-peak, including weekends. Off-peak prices are half of peak prices.
Does DeepSeek work with the OpenAI or Anthropic format?
Yes. The base URL for the OpenAI format is https://api.deepseek.com. The base URL for the Anthropic format is https://api.deepseek.com/anthropic. Developers can use compatible SDKs or tools by changing their settings.
