Alibaba Presents Qwen-3: A New AI With a Trillion Parameters

Chinese tech giant Alibaba has unveiled Qwen-3-Max-Preview, a text-based artificial intelligence model with over 1 trillion parameters. It significantly outperforms the company's previous most powerful model, Qwen3-235B-A22B-2507, released in July this year.

Competition With Global Players

According to benchmark results, Qwen-3-Max-Preview outperformed MoonShot AI's Kimi K2, Anthropic's Claude Opus 4 (without logical thinking), and DeepSeek V3.1 in a series of tests. However, the results have not yet been confirmed by an official technical report.

For comparison, one of the largest models in the world, GPT-4.5 from OpenAI, likely contains between 5 and 7 trillion parameters.

Qwen's Popularity in the Community

The Qwen series of models has made Alibaba a leader in open AI solutions. They have been downloaded more than 20 million times on the Hugging Face platform, creating about 100,000 derivative versions. Despite this, the new Qwen-3-Max-Preview does not yet have open source code, just like the previous Qwen2.5-Max.

Cost of Use and Prospects

Qwen-3-Max-Preview has one of the highest prices in the series: $0.861 per million input tokens and $3.441 per million output tokens. For comparison, the previous model costs $0.287 and $1,147, respectively, and Kimi K2 costs $0.6 and $2.5.

The company also announced an upcoming version of Qwen-3-Max-Preview with logical thinking capabilities.

Infrastructure Investment

Alibaba plans to invest $52 billion in AI infrastructure development over three years. In addition, according to media reports, the company is working on its own artificial intelligence processor to reduce its dependence on Nvidia.