Alibaba, the Chinese tech giant, has launched eight new open-weight AI models as part of the Qwen 3 series. The new models include two MoE (Mixture of Experts) variants: Qwen3-235B-A22B and Qwen3-30B-A3B. Qwen3-235B-A22B stands as the flagship model, featuring 235 billion parameters with 22 billion activated parameters.
On the other hand, Qwen3-30B-A3B is a more compact MoE model, equipped with 30 billion parameters and 3 billion activated parameters. In addition, the Qwen 3 series also offers six dense models: Qwen3-32B, Qwen3-14B, Qwen3-8B, Qwen3-4B, Qwen3-1.7B, and Qwen3-0.6B.
All Qwen 3 models feature Hybrid Thinking Modes, combining reasoning AI capabilities with traditional LLM functionalities. In the Thinking mode, the model can reason through tasks step by step, while in the Non-Thinking mode, it provides quick responses.
Additionally, Qwen 3 models support over 119 languages and dialects from across the globe, making them some of the most diverse multilingual models available. Alibaba has also enhanced MCP support for Qwen 3 models, further unlocking agentic capabilities.
In terms of performance, the Qwen3-235B-A22B model stands out, delivering competitive results comparable to DeepSeek R1, Grok 3 Beta, Gemini 2.5 Pro, and OpenAI’s o1. Interestingly, the smaller Qwen3-30B-A3B model, with only 3 billion activated parameters, outperforms both DeepSeek V3 and OpenAI’s GPT-4o model.
Alibaba claims that Qwen 3 models excel in areas like coding, math, science, and general capabilities. Overall, Qwen 3 represents a powerful family of advanced AI models from China. With the upcoming DeepSeek R2, China is positioned to effectively compete with leading Western AI labs.
