DeepSeek
DeepSeek is an open-weight AI assistant with a free consumer chat app, a 1M-token context window, and a low-cost API supporting thinking mode and agent workflows.
What is DeepSeek?
DeepSeek is an AI assistant developed by Hangzhou DeepSeek Artificial Intelligence. Its product structure has three distinct layers that are often conflated: a free consumer chat application at chat.deepseek.com and on mobile, a paid API for developers, and open model weights released under the MIT license that allow local deployment and commercial use.[reference:0] These layers share model technology but differ in access, cost, and intended audience.
The current generation is the DeepSeek-V4 family, comprising deepseek-flash and deepseek-v4-pro. Both models support a 1M-token context window, a maximum output of 384K tokens, and a thinking mode that can be toggled and tuned via a reasoning_effort parameter. deepseek-flash additionally supports image understanding; deepseek-v4-pro does not. The consumer web interface offers Fast Mode and Expert Mode as interaction layers, with different functional boundaries.
How does DeepSeek work?
In the consumer chat application, users type natural-language instructions or, in Fast Mode, upload files and images for text recognition. The interface provides toggles for deep thinking and web search. Expert Mode, positioned for complex reasoning, supports deeper thinking and intelligent search but does not currently accept file uploads or multimodal input.
Through the API, developers send requests in OpenAI-compatible or Anthropic-compatible formats to either deepseek-flash or deepseek-v4-pro. The thinking mode is enabled by default and can be controlled with thinking and reasoning_effort parameters. For multi-turn conversations with tool calls, the reasoning_content from prior turns should be returned to the API. Uploaded images via the Files API are referenced by file_id and are used together with deepseek-flash. Outputs can be returned as text, JSON, or structured tool calls, depending on the request configuration.
Use Cases
- Drafting and editing long documents where the 1M-token context window reduces the need to chunk or summarize source material
- Analyzing and extracting information from uploaded images or scanned text in the consumer Fast Mode or via the API with deepseek-flash
- Working through multi-step reasoning, math, or STEM problems with thinking mode enabled
- Building agentic coding or automation workflows through OpenAI- or Anthropic-compatible APIs, with concurrency limits differing between Flash and Pro
- Deploying the model locally or within an organization using the MIT-licensed open weights, without paying per-token API fees
Pricing
Pricing model: Freemium
| Plan | Price | Billing | Included usage | Notes |
|---|---|---|---|---|
| Free (Chat App) | Free | — | Web and mobile chat access with no subscription, no paid consumer tier, and no published daily message quota. Server capacity during peak periods is the primary practical constraint. | — |
| API (Pay-as-you-go) | Variable | — | Billed per 1M tokens. Off-peak rates are half of peak rates. deepseek-flash: $0.15 input (cache miss) / $0.60 output off-peak; deepseek-v4-pro: $0.66 input (cache miss) / $1.98 output off-peak. Search and Fetch APIs are billed separately per successful request. | — |
Common Limitations
- Consumer access model · The chat application has no paid consumer plan, so there is no higher-tier consumer subscription to unlock additional model access or higher message quotas. Access is governed by server capacity rather than a published daily or hourly message cap.
- Peak-hour availability · During peak periods, users may encounter slower generation or 'server busy' responses. The Expert Mode in the web interface explicitly notes that users may need to wait during peak hours.
- Expert Mode functional boundary · The web interface's Expert Mode, designed for complex problems, does not currently support file uploads or multimodal recognition. File text recognition is available in Fast Mode.
- Model-level capability differences · deepseek-flash supports image understanding (Vision), while deepseek-v4-pro does not. FIM Completion (Beta) is only supported in non-thinking mode for both models.
- File upload constraints (API) · Files uploaded via the Files API are limited to 64 MiB per file and must complete upload within 10 minutes. If no expiration is set, files are stored permanently. These limits apply to the API and are distinct from any limits in the web chat interface.
- API billing scope · API usage is billed separately from the free chat application. Search and Fetch are standalone APIs billed per successful request, in addition to model token usage.
Key Features
- 1M-token context window across V4 models
- Thinking mode with adjustable reasoning_effort (low/high/max)
- OpenAI- and Anthropic-compatible API
- Free consumer chat app without a paid subscription tier
- Image understanding on deepseek-flash only
DeepSeek vs. ChatGPT, Claude, and Gemini
This comparison draws on current official product and pricing pages. It focuses on documented positioning, access structure, and capability boundaries rather than subjective rankings of model quality. Pricing and feature availability can change and should be verified against each provider's current documentation.
| Tool | Positioning | Key differences | Best fit | Pricing note |
|---|---|---|---|---|
| DeepSeek Official product overview API models & pricing V4 preview announcement | An AI assistant built around a free consumer chat app, a pay-as-you-go API, and MIT-licensed open weights. | Offers a 1M-token context window across its V4 model family, a consumer chat app with no paid subscription tier, and open weights for local or commercial deployment. API pricing is structured with off-peak discounts. | Users or developers who prioritize a free consumer experience, long-context tasks, low API costs, or the ability to self-host open-weight models. | Consumer chat is free with no paid tier. API is pay-as-you-go; off-peak rates are half of peak. No permanent free API tier. |
| ChatGPT Official overview Pricing | A general-purpose assistant for chatting, working, and coding. | Combines writing, file and data analysis, web search, Deep Research, image generation, coding with Codex, voice, and memory. Access to models and tools varies by plan. Does not offer open model weights. | Users who want a broad set of multimodal, research, coding, and work tools in one product with a mature paid subscription structure. | A Free plan is available. ChatGPT Plus is $20 per month; API usage is billed separately. |
| Claude Official overview Pricing | An AI assistant for work, emphasizing deliverables, artifacts, and coding. | Anthropic highlights creating and scheduling deliverables such as decks, spreadsheets, and prototypes through Artifacts, coding with Claude Code, and connecting apps and tools. The product structure is subscription-oriented for consumers. | Workflows centered on document and file deliverables, prototypes, task handoffs, and coding. | A Free plan is available. Claude Pro is $20 per month or $200 when billed annually. |
| Gemini Official overview | Google's multimodal and personal AI assistant. | Google highlights multimodal input, document summarization, coding, writing, image generation, Deep Research, and connections to Google products and services. Consumer access is structured around Google AI plans. | Users who want a multimodal assistant closely integrated with Google's ecosystem and services. | Free access and paid Google AI plans are available; regional pricing and availability vary. |
Comparison checked against official product information on 2026-11-04. Features and pricing may change.
Frequently Asked Questions
Is DeepSeek free to use?
The consumer chat application (web and mobile) is free, with no subscription, paid tier, or published daily message quota. The API is not free; it is billed per token on a pay-as-you-go basis. Open weights are available under the MIT license for local deployment, which shifts costs to infrastructure rather than API fees.
Does DeepSeek have a paid consumer plan like ChatGPT Plus?
No. As of the current documentation, DeepSeek's consumer chat application does not have a paid subscription tier. The web interface offers Fast Mode and Expert Mode, but these are interaction modes rather than paid plan levels. Paid access exists through the API, not through a consumer subscription.
What is the difference between deepseek-flash and deepseek-v4-pro?
Both models support a 1M context window, 384K max output, thinking mode, and OpenAI/Anthropic-compatible APIs. deepseek-flash supports image understanding and has a higher concurrency limit (2500 vs 500). deepseek-v4-pro is priced higher and does not support Vision. deepseek-flash is the more economical option for tasks where its knowledge and high-end reasoning are sufficient.
What are the file upload limits?
For the Files API, a single file may be at most 64 MiB and the upload must complete within 10 minutes. Supported formats are JPEG, PNG, GIF, and WebP. Files can be set to expire between 1 hour and 30 days, or kept permanently if no expiration is specified. Web chat file limits are displayed by the interface and may differ from API limits.
Does the API include free usage?
No. DeepSeek's API is pay-as-you-go with no permanent free tier. Off-peak rates are half of peak rates. Peak hours are 01:00–04:00 and 06:00–10:00 UTC, Monday through Friday, excluding Chinese public holidays.
Alternative Tools
-
Qwen Chat Qwen Chat is Alibaba's free consumer AI assistant, offering 1M-token context, multimodal understanding, image and video generation, Deep Research, and web development tools, with optional paid consumer tiers for higher quotas. -
ChatGPT ChatGPT is a general-purpose AI assistant for writing, file and data analysis, web research, image generation, coding, and voice conversations. -
Claude Claude is Anthropic's AI assistant for writing, analysis, coding, web search, and delegating tasks that return finished documents, slides, spreadsheets, and designs. -
Grok Grok is xAI's AI assistant for chat, real-time search, image and video generation with Grok Imagine, voice conversations, file analysis, and coding and agentic tasks through Grok 4.5. -
Google Gemini Google Gemini is Google's multimodal AI assistant, offering up to a 1M-token context window on paid plans, Deep Research, and deep integration with Google apps and services, with a Free tier now limited to the Flash-Lite model. -
Vibe (formerly Le Chat) Vibe (formerly Le Chat) is Mistral AI's unified assistant for chat, multi-step work tasks with connected apps, and agentic coding across web, IDE, and terminal, with a Free tier and paid Pro and Team plans.