General chat
DeepSeek-V series
Best for: Everyday chat, writing, coding
Large mixture-of-experts models; the default model in the app.
🇨🇳 China · China
Hangzhou-based DeepSeek became famous for high-performing, low-cost models released with open weights. Its models power the free DeepSeek chat app and a low-price API.
General chat
Best for: Everyday chat, writing, coding
Large mixture-of-experts models; the default model in the app.
Reasoning
Best for: Maths, logic, step-by-step problems
Shows its reasoning; used in the app’s “DeepThink” mode.
Small open reasoning models
Best for: Running reasoning locally on a laptop
Smaller models (based on Qwen and Llama) trained to reason like R1 — e.g. ollama run deepseek-r1:7b.
Code
Best for: Code generation and completion
Code-specialised models.
Model names and versions change often. Check the official documentation for the newest releases.
Cloud services from Chinese companies process data under Chinese law and the provider’s privacy policy. Many organisations in the UK, EU and US restrict their use for sensitive data. The open-weight models can instead be run locally or on your own servers, where your data stays under your control.
The open-weight models run fully on your computer when used locally. For the cloud app, your data is processed in China under DeepSeek’s policy — avoid sensitive information and follow your organisation’s rules.
Yes — the smaller distilled versions run on ordinary laptops with Ollama or LM Studio. See our step-by-step guide.