The Qwen team releases Qwen3, a new family of open-weight large language models that achieve top-tier benchmark results in coding and math. The release includes efficient Mixture-of-Experts models and smaller dense models available under the Apache 2.0 license.