DeepSeek
by DeepSeek (Hangzhou DeepSeek Artificial Intelligence)
Open-weight LLMs plus a free chat assistant and a low-cost OpenAI-compatible API
Last reviewed 2026-06-20
DeepSeek is a Chinese AI research lab that builds large language models and ships them three ways: as open-weight model releases under the MIT license, as a free consumer chat assistant (web and mobile app), and as a low-cost developer API that is OpenAI- and Anthropic-compatible. Its models include the DeepSeek-V3 and DeepSeek-R1 generation and the newer DeepSeek-V4 family (V4-Flash and V4-Pro), with a separate thinking/reasoning mode for harder problems. The chat product is a general-purpose assistant: it answers questions, writes and explains code, reads uploaded files, and holds long-context conversations. It responds when asked rather than taking independent action, so it sits at the assistant end of the autonomy ladder. DeepSeek drew global attention in January 2025 when its R1 reasoning model, released as open weights and reportedly trained far more cheaply than US frontier models, briefly topped the US iOS App Store and rattled AI-chip stocks. It targets consumers for free chat and developers/businesses via the API and self-hostable weights.
What it can do
General conversational assistant
AssistantFree web and mobile chat assistant that answers questions, writes content, reads uploaded files, and holds long-context conversations; it responds on request and does not take independent action.
sourceCode generation and explanation
CopilotWrites, debugs, and explains code; the models are used as a coding backend in tools like Claude Code and GitHub Copilot via the OpenAI/Anthropic-compatible API.
sourceReasoning / thinking mode
AssistantA chain-of-thought reasoning mode (the R1 lineage, now the thinking mode of DeepSeek-V4) for harder math, logic, and multi-step problems, configurable via a thinking parameter and adjustable reasoning effort.
sourceOpenAI-compatible developer API with tool calls
AssistantA chat-completions API that is compatible with the OpenAI and Anthropic SDKs (point the base URL at api.deepseek.com) and supports tool/function calling and JSON output, so developers can build agentic apps on top of it.
sourceOpen-weight model releases
AssistantReleases model weights (DeepSeek-V3, R1, and the V4 family) under the MIT license, so they can be downloaded, self-hosted, fine-tuned, and run by third-party providers.
source
Strengths
- +Free consumer chat assistant and a notably low-cost API versus US frontier providers
- +Open-weight models under the MIT license, so they can be self-hosted, fine-tuned, and run by third parties
- +OpenAI- and Anthropic-compatible API makes it a near drop-in for existing apps and coding tools
Limitations
- −It is an assistant, not an autonomous agent: it responds when asked and does not act end-to-end
- −China-hosted service raises data-residency and privacy concerns, and the app has faced government bans and scrutiny in several countries
- −Reasoning mode can consume many internal reasoning tokens, eroding the headline price advantage on hard tasks
Overview
DeepSeek is a Chinese AI lab (founded 2023, Hangzhou, backed by the hedge fund High-Flyer) that builds large language models and ships them as open-weight releases, a free chat assistant, and a low-cost API. It became globally known in January 2025 when its R1 reasoning model, released as open weights and reportedly trained far more cheaply than US frontier models, briefly topped the US iOS App Store and pressured AI-chip stocks (Wikipedia).
What it does
The chat assistant answers questions, writes and explains code, reads uploaded files, and holds long-context conversations. It responds when asked and does not act on its own, so it is an assistant, with copilot-grade coding help. A separate thinking/reasoning mode (the R1 lineage, now the thinking mode of the V4 family) targets harder math, logic, and multi-step problems. The model family spans DeepSeek-V3, R1, and the newer V4-Flash and V4-Pro.
Integrations & setup
The developer API is a chat-completions endpoint that is compatible with the OpenAI and Anthropic SDKs: point the base URL at api.deepseek.com and existing code works. It supports tool/function calling and JSON output, and DeepSeek documents using it as the backend for tools like Claude Code and GitHub Copilot. Because the weights are MIT-licensed and open, the models can also be self-hosted or accessed through third-party inference providers.
Pricing
Freemium. The web and mobile chat assistant is free. The API is usage-based and inexpensive: per the official pricing page, DeepSeek-V4-Flash is roughly $0.14 per 1M input tokens (cache miss) and $0.28 per 1M output, with V4-Pro higher and cache hits an order of magnitude cheaper. Self-hosting the open weights is free aside from compute.
Best for / not for
Best for developers who want a cheap, OpenAI-compatible model API or self-hostable open weights, and consumers who want a free, capable chat and coding assistant. Less suited to teams that need an autonomous agent that takes actions end-to-end, or to organizations with strict data-residency or compliance constraints given the China-hosted service and the bans it has faced in several jurisdictions.
Alternatives
ChatGPT, Google Gemini, Perplexity, and Microsoft Copilot are the mainstream conversational-assistant alternatives. Versus those, DeepSeek's differentiators are open weights and aggressively low API pricing; its tradeoffs are autonomy (it is an assistant, not an agent) and the privacy and governance questions around a China-hosted provider.
What people are saying
We aggregate real LinkedIn discussion into sentiment for the agents people search most. DeepSeek isn't tracked yet, want it added? Request tracking.
FAQ
Is DeepSeek an autonomous agent?+
No. DeepSeek is a conversational assistant and a set of open-weight LLMs. It answers, writes, and reasons when prompted, and its coding help is copilot-grade, but it does not take multi-step actions on your behalf without you driving it. Developers can build agents on top of its API (it supports tool/function calling), but the product itself is an assistant.
Are DeepSeek's models open source?+
The weights are open under the MIT license (open-weight), so anyone can download, self-host, fine-tune, and serve them. The training data and full training pipeline are not released, so it is open-weight rather than fully open-source.
How much does DeepSeek cost?+
The chat assistant is free. The API is usage-based and low-cost: per the official pricing page, DeepSeek-V4-Flash is about $0.14 per 1M input tokens (cache miss) and $0.28 per 1M output, with V4-Pro higher; cache hits are far cheaper. You can also self-host the open weights for free (compute aside).
Sources
- DeepSeek chat assistant (official) · accessed 2026-06-20
- DeepSeek API documentation and overview · accessed 2026-06-20
- DeepSeek API pricing (official) · accessed 2026-06-20
- DeepSeek on GitHub (open-weight model releases) · accessed 2026-06-20
- DeepSeek (Wikipedia): founding, models, January 2025 R1 launch and market impact · accessed 2026-06-20
Last reviewed 2026-06-20