The Qwen3 family consists of eight models, ranging from 600 million parameters to 235 billion, with enhancements across all models, according to the Qwen team at Alibaba’s cloud computing unit. Alibaba owns the South China Morning Post.
In AI, parameters are a measurement of the variables present during model training. They serve as an indicator of sophistication: larger parameter sizes typically suggest greater capacity.
Benchmark tests cited by Alibaba revealed that models such as Qwen3-235B and Qwen3-4B matched or exceeded the performance of advanced models from both domestic and overseas competitors – including OpenAI’s o1, Google’s Gemini and DeepSeek’s R1 – in areas like instruction following, coding assistance, text generation, mathematical skills and complex problem solving.