Qwen3
A local LLM family with public weights, chat and reasoning variants, and CPU or GPU support through Ollama, llama.cpp, and LM Studio.
Qwen3 is a language model family from Alibaba Cloud's Qwen team for people building chat applications or running models on their own hardware. Its public weights include dense and mixture-of-experts models in smaller and larger sizes.
- Chat and writing: Instruct models generate direct responses for conversation, text generation, and instruction following, with support for multilingual text.
- Reasoning: Thinking models handle problems in logic, mathematics, science, and coding. Dedicated Instruct and Thinking variants each use a single response mode; other Qwen3 models can switch between thinking and direct answers.
- Local inference: Run models on CPU or GPU through llama.cpp, Ollama, or LM Studio, or use Transformers for Python applications.
- Server deployment: vLLM, SGLang, and TGI support serving the models on your own infrastructure. Through llama.cpp, applications can access an OpenAI-compatible API.
Qwen3 supports GGUF, GPTQ, and AWQ quantization as alternatives to full-size weights. Model checkpoints are available through Hugging Face and ModelScope.
Logotyper
