Foreign media: China's Alibaba has released the largest artificial intelligence model to date, Qwen-3-Max-Preview, with parameter count exceeding 1 trillion, officially entering the same level of competition as industry giants such as OpenAI and Google's DeepMind.
This is the first time Alibaba has launched a model with over 1 trillion parameters, and it has been launched on Alibaba Cloud and the large model platform OpenRouter. The model is the latest member of the Qwen3 series launched in May this year. Previous versions had parameter scales ranging from 600 million to 235 billion. Alibaba claims that the new model has seen "significant improvements" in understanding Chinese and English, executing complex instructions, and handling open-ended subjective tasks. In internal testing, it performed better than the previous Qwen3-235B-A22B-2507, and it also surpassed MoonShot AI's Kimi K2, Anthropic's Claude Opus 4 non-inference version, and DeepSeek V3.1 in five benchmark tests, but no official technical report has been released yet.
In general, the more parameters a model has, the stronger its capabilities are, but the computational power required for training and operation also increases. Currently, OpenAI's GPT-4.5 is estimated to have between 5 and 7 trillion parameters.
Original article: www.toutiao.com/article/1842706157999108/
Statement: This article represents the personal views of the author.