Alibaba Releases Qwen3 Series with Hybrid Thinking and 235B MoE Model

Alibaba's Qwen team released the Qwen3 series, including dense models from 0.6B to 32B parameters and a 235B Mixture-of-Experts flagship, all available as open weights. Qwen3 introduces hybrid thinking mode allowing models to switch between fast response and extended chain-of-thought reasoning. The 32B dense model claims to outperform DeepSeek-R1 and o1 on math and coding benchmarks while supporting 119 languages.