Alibaba's Qwen3.8-Max preview model reaches 2.4 trillion parameters

Alibaba's Tongyi Qianwen team has released the preview version of Qwen3.8-Max, a massive AI model with 2.4 trillion parameters, according to Chinese media reports.

The new model extends the Qwen series architecture to unprecedented scale. At 2.4 trillion parameters, Qwen3.8-Max places Alibaba's foundation model in direct competition with the world's largest models from Google Gemini, OpenAI, and other frontier labs.

Ultra-large parameter models generally offer greater knowledge capacity and more sophisticated reasoning capabilities, but they also demand enormous training compute and pose significant inference deployment challenges. Alibaba's decision to release a Max preview suggests the team has made architectural or training breakthroughs.

The preview launch comes amid intensifying competition in China's foundation model landscape. Baidu's Ernie, ByteDance's Doubao, and Huawei's Pangu are all iterating rapidly, and Alibaba's Qwen3.8-Max preview signals its determination to maintain a technological edge.

Detailed benchmark evaluations for the preview version have not yet been fully published, but the industry is closely watching its performance on math reasoning, code generation, and multilingual comprehension — areas where previous Qwen models have performed well in the open-source community.

With parameter counts now exceeding the trillion mark, inference cost and deployment efficiency become the critical bottlenecks for real-world adoption. Alibaba Cloud's infrastructure and commercial reach will determine whether Qwen3.8-Max can move from preview to widespread production deployment.