DeepSeek
Chat assistant and open language models from China, usable via web, app and a low-cost API, or self-hosted under the MIT licence.
What it does
DeepSeek is a chat assistant built on its own language models, which you use in the browser, via an app or through an API. The current model, V4.1-Flash, handles text and images, has a thinking mode for step-by-step reasoning and can call tools. The context window holds up to one million tokens. The model weights are published on Hugging Face under the MIT licence, so you can run them on your own hardware. DeepSeek Harness, an open-source agent for macOS and Windows, analyses documents, writes code and fixes bugs in projects.
Who it suits
Development teams that want low-cost API access to capable models, and companies with their own GPU infrastructure that want to self-host an open model. In everyday work, use the chat only for non-sensitive content.
Data protection for businesses
DeepSeek stores data from chat and app in China. The vendor does not offer EU data hosting. Inputs are used for training by default; you can object by emailing privacy@deepseek.com. The vendor does not mention a data processing agreement or business plans. If you need to comply with the GDPR, run the open models yourself or with a European hosting provider.
Limits
Self-hosting the large models requires substantial computing power. During weekday peak hours the API costs twice as much. Answers can be factually wrong and need checking.