The main names
- DeepSeek (by DeepSeek AI): known for strong reasoning, maths and coding. Offers a free web chat, mobile apps, a low-cost API, and open-weight models you can run locally.
- Qwen (by Alibaba Cloud): a large family of models in many sizes — from small models that run on a laptop to large ones in the cloud — plus vision and coding versions. Available through Qwen Chat and as open weights.
- Kimi (by Moonshot AI): known for handling very long documents and for capable agent-style models. Used mainly through the Kimi website and app.
- Others you may see: GLM / ChatGLM (Zhipu AI), Doubao (ByteDance), ERNIE (Baidu), Hunyuan (Tencent) and MiniMax.
What “open-weight” means for you
An open-weight model publishes the trained model files. Anyone can download them and run the model with free software such as Ollama or LM Studio. When you run a model locally, your prompts stay on your computer and it works without internet after the download.
The trade-off is hardware: small and medium versions run on an ordinary laptop, but the largest versions need powerful graphics cards far beyond a home PC. Locally you will usually run a smaller “distilled” or reduced version, which is capable but not as strong as the full cloud model.
Three ways to use them
- 1
Web or mobile app
The easiest start: open the official website or install the official app and sign in.
- 2
Run locally
Install Ollama or LM Studio and download an open-weight model. Private and offline, limited by your hardware.
- 3
API for developers
Create an API key on the official developer platform and call the model from your own code.
Privacy: what to know
When you use any cloud AI service, your prompts are processed on that company’s servers under their privacy policy and the laws of the country where the data is handled. Read the policy, avoid sharing sensitive personal or company data, and follow your employer’s rules. If privacy is the priority, running an open-weight model locally keeps your data on your own device.