DeepSeek
** - A powerful AI platform offering **open-source LLMs** such as **DeepSeek-V3 (671B)** and **DeepSeek R1 (660B)**, along with **smaller models (1.5B, 7B, 14B)** for local deployment. It also features a **paid API** compatible with OpenAI's format, supporting **streaming and enterprise solutions**.
Last verified:
What is DeepSeek?
DeepSeek is an AI assistant and large language model platform that offers chat, API access, and open-source models for developers, researchers, and general users. The platform provides DeepSeek-V4 (Preview), a world-class model with advanced reasoning capabilities, enhanced Agent performance for coding tasks, and 1M context length support. Users can access DeepSeek through the free web chat interface, mobile app, or API for programmatic integration.
Key features include Deep-Think mode for complex reasoning, web search integration, file upload support (PDF, DOCX, TXT up to 10MB), JSON output for structured data, Function Calling for tool interaction, Chat Prefix Completion, and FIM (Fill-In-the-Middle) Completion for code/story completion. The API supports streaming responses, is OpenAI-compatible, and offers both thinking and non-thinking modes. DeepSeek also provides open-source models like DeepSeek R1, V3, Coder V2, and VL for local deployment.
DeepSeek is designed for multiple user groups: general users seeking a free AI chatbot for everyday tasks, developers integrating AI into applications via API, researchers exploring open-source LLMs, and coding professionals needing agentic coding assistance. The platform serves both consumer and enterprise needs with its combination of free consumer access and metered developer API pricing.
DeepSeek pricing
Pricing model: Freemium
DeepSeek offers a fully free consumer tier with no subscription required: the web chat interface and mobile app are 100% free with all features including web search, Deep-Think mode, file upload, text extraction, and chat history sync. For developers, the API uses token-based billing with no free tier but trial credits granted upon sign-up. Current API pricing (April 2026): DeepSeek V3.2 (deepseek-chat/deepseek-reasoner) costs $0.14 per 1M input tokens and $0.28 per 1M output tokens, with cached input at $0.014 per 1M tokens (90% discount). DeepSeek-V4-Flash costs $0.14/1M input (cache hit: $0.0028) and $0.28/1M output. DeepSeek-V4-Pro costs $0.435/1M input during 75% promotion (cache hit: $0.003625) and $0.87/1M output, with prices adjusting to 1/4 original after May 31, 2026. Users top up via PayPal, bank card, Alipay, or WeChat Pay; balances don't expire and unused amounts are refundable.
DeepSeek pros
- 1M context length as standard across all services
- Completely free consumer chat and mobile app with no subscription tiers
- World-class reasoning performance rivaling top closed-source models
- Enhanced agentic coding capabilities with Open-source SOTA benchmarks
- Extremely affordable API pricing at $0.14 per 1M input tokens
- 90% cache discount reducing cached input to $0.014 per 1M tokens
- OpenAI-compatible API for easy integration with existing tools
- Both thinking and non-thinking modes available for flexibility
- JSON Output feature ensures valid structured JSON responses
- Function Calling enables interaction with external tools
- Open-source models available for local deployment and commercial use
- Web search integrated directly into chat interface
- File upload support with text extraction capabilities
- Streaming response support for better interactivity
- 384K maximum output tokens for long-form generation
- Chat history synchronization across devices with account
DeepSeek cons
- API occasionally returns empty content requiring prompt adjustments
- Rate limits and concurrency caps during high-demand periods
- Throughput affected by server outages and platform traffic
- Some email domains not supported for registration
- Account suspension possible if usage guidelines are triggered
- Free tier has no official message caps but limited by capacity
- Commercial use restricted for free accounts (requires paid API)
- Fine-tuning not available on free plan
- deepseek-chat and deepseek-reasoner deprecated after July 24, 2026
- 75% discount promotion ends May 31, 2026 with price adjustment
Frequently asked questions about DeepSeek
What is DeepSeek-V4 and what makes it different from previous versions?
DeepSeek-V4 is the latest flagship model with two variants: V4-Pro (1.6T total parameters, 49B active) for deep reasoning and agentic coding, and V4-Flash (284B total, 13B active) for fast, cost-effective usage. Key innovations include 1M standard context length, token-wise compression with DeepSeek Sparse Attention, three reasoning effort modes, and enhanced agentic capabilities seamlessly integrated with Claude Code, OpenClaw, and OpenCode. V4 beats all open models in Math/STEM/Coding and rivals top closed-source models in reasoning.
Is DeepSeek free to use?
Yes, DeepSeek's consumer chat and mobile app are completely free with no subscription tiers, ads, or paywalled upgrades. All major features including web search, Deep-Think mode, file upload, text extraction, and chat history sync are available to free users. The API is metered through token-based billing with trial credits granted upon sign-up, but there are no paid subscription plans for consumer access.
How do I get an API key for DeepSeek?
Visit the DeepSeek website and apply for an API key through the Open Platform. After registering with a supported email (Gmail, Outlook, Hotmail, or Yahoo recommended), you'll receive a free API key with trial credits. The key is required for all API requests and should be kept secure. You can top up your balance via PayPal, bank card, Alipay, or WeChat Pay on the Top Up page.
What is the DeepSeek API base URL and is it compatible with OpenAI?
The DeepSeek API base URL is https://api.deepseek.com, with OpenAI-compatible format at https://api.deepseek.com/v1. The API is designed to be compatible with OpenAI's API, so you can use OpenAI's SDKs and libraries by simply changing the base_url and providing your DeepSeek API key. It also supports Anthropic API format at https://api.deepseek.com/anthropic.
What is the difference between thinking mode and non-thinking mode?
Thinking mode enables DeepSeek's deep reasoning capabilities, allowing the model to work through complex problems step-by-step before answering. Non-thinking mode provides faster responses for simpler tasks. Both modes are available for deepseek-v4-flash and deepseek-v4-pro, with thinking as the default. You can switch between modes in your API calls or chat interface. The old deepseek-chat corresponds to non-thinking mode and deepseek-reasoner corresponds to thinking mode of deepseek-v4-flash.
How do I enable JSON Output in the DeepSeek API?
To enable JSON Output: 1) Set the response_format parameter to {'type': 'json_object'}, 2) Include the word 'json' in your system or user prompt and provide an example of the desired JSON format, 3) Set max_tokens reasonably to prevent truncation. The API enforces valid JSON output, which is useful for data processing and automation. Note that the API may occasionally return empty content when using JSON Output, and modifying prompts can help mitigate this.
What are the rate limits for DeepSeek API?
Concurrency limits vary by model: deepseek-v4-flash has 2500 concurrent requests, while deepseek-v4-pro has 500. Free accounts typically have 10-30 requests per minute with 1-3 concurrent requests. Rate limits and capacity caps apply to ensure fair use, and throughput may be affected during high-demand periods or server outages. For capacity expansion requests, refer to the Rate Limit & Isolation documentation or contact [email protected].
Can I download and run DeepSeek models locally?
Yes, DeepSeek provides open-source weights and inference code for local deployment. Models like DeepSeek R1, V3, Coder V2, and VL are available for download with no cloud quota when running on personal hardware. Open-source weights are MIT licensed, allowing free distillation and commercial use. Local inference is unlimited for research purposes, though hosted cloud APIs enforce quotas. You can pull models using tools like Ollama (e.g., 'ollama pull deepseek-r1').
What file types can I upload to DeepSeek and what are the limits?
DeepSeek supports file uploads in the web playground and via API for PDF, DOCX, and TXT formats. The maximum file size is 10 MB per file, with 1-3 files allowed per session. Built-in text extraction enables the model to read and process content from uploaded files. File upload is available to free users without subscription requirements.
Why is my account suspended and how can I appeal?
Account suspension occurs when the account triggers suspension protocols due to potential violations of platform usage guidelines. If you see 'your account has been temporarily suspended' when logging in, you can appeal by filling out the Account Suspension Appeal form. Most reviews are completed within 3 business days, and once approved, you can log in and use the service immediately. If you believe the suspension is an error, contact [email protected].