Qwen 3 is an open model in the Qwen series.There are multiple releases available including a 30/3 billion parameter MoE model.
Qwen3 models are hybrid reasoning models that can be prompted to think or not by passing special tokens /think or /no_think in the prompt. Thinking is apparently enabled by default.
Qwen3 can be easily run within ollama and are supported in v0.8.5 of vllm