DeepSeek

What it's good for

• 1

  Free everyday chat without a subscription

• 2

  Adding AI features to a product at a fraction of the usual API cost

• 3

  Self-hosting open weights when data must stay on your own infrastructure

• 4

  High-volume batch jobs (classification, extraction) where per-token cost dominates

• 5

  Prototyping against a cheap API before committing to a pricier frontier model

Pricing

• Chat

  Web assistant, no paid tier

  Free

• API V4 Flash

  $0.28 per M output tokens

  $0.14/M in

• API V4 Pro

  $3.48 per M output tokens

  $1.74/M in

• Self-hosted

  Open weights, MIT license

  Free

How to use it

Use the web chat at chat.deepseek.com for free. For product work, grab an API key and point your existing client at the DeepSeek endpoint (it speaks the standard chat-completions shape, so most SDKs work with a base-URL swap). When cost at scale or data control becomes the deciding factor, pull the open weights and run them on your own hardware or a GPU host.

Pros & cons

Frequently asked questions

• Is DeepSeek free?

  The web chat is free with no paid tier, and it competes with the paid frontier assistants for everyday use. The API is paid but very cheap: V4 Flash is $0.14 in / $0.28 out per million tokens and V4 Pro is $1.74 / $3.48. The open weights are free to self-host under an MIT license.

• DeepSeek vs ChatGPT: what is the difference?

  ChatGPT is the broader, more polished product with deep tool integration and a large ecosystem. DeepSeek's draw is cost and openness: a free web chat plus an API priced low enough to change the math on AI features, with open weights you can run yourself. If price-per-token or data control drives the decision, DeepSeek wins.

• Is DeepSeek open source?

  The model weights are open under an MIT license, so you can download and self-host them when data must stay on your own infrastructure or when cost at scale matters. That is a big difference from closed frontier models you can only reach through a vendor API.

• Can I self-host DeepSeek?

  Yes. Because the weights are MIT-licensed, you can run them on your own hardware or a GPU host with no per-token vendor fee. That makes it a fit for high-volume jobs and for teams that need code or data to stay in their own environment.

• What is DeepSeek best for?

  Free everyday chat, adding AI features to a product cheaply, and high-volume token jobs like classification or extraction where per-token cost dominates. As with any hosted model, route sensitive data deliberately and read the terms before piping production traffic through the endpoint.

More

• DeepSeek API docs

Alternatives to DeepSeek

Other tools we'd consider for the same job.

• Claude

  Anthropic's conversational AI — strong at careful reasoning, long documents, and writing that reads like a human wrote it

• ChatGPT

  OpenAI's assistant. Deep tool integration, Advanced Voice, image generation, and the default for most first-time AI users

• Mistral Le Chat

  EU-sovereign frontier assistant with image gen, a code interpreter, and 40+ connectors. Built for data-residency rules

Other Chat & Assistants tools

• Claude

  Anthropic's conversational AI — strong at careful reasoning, long documents, and writing that reads like a human wrote it

• Mistral Le Chat

  EU-sovereign frontier assistant with image gen, a code interpreter, and 40+ connectors. Built for data-residency rules

• ChatGPT

  OpenAI's assistant. Deep tool integration, Advanced Voice, image generation, and the default for most first-time AI users

• Gemini

  Google's assistant with 1M-token context, native Google Workspace integration, and strong multimodal reasoning