Alibaba's Qwen3.8 joins world's top-tier LLMs

Alibaba Group has unveiled its latest 2.4 trillion-parameter AI model Qwen3.8, placing it among the world's top-tier large language models in overall performance.

Alibaba said Qwen3.8 trails only Anthropic's Claude models on Arena, a widely followed third-party AI evaluation platform.

Optimized for coding and "cowork" scenarios, Qwen3.8 allows users to apply AI more directly to software development, document processing, data analysis and other professional workflows.

Alibaba plans to open-source Qwen3.8-Max next week, alongside Qwen3.8-27B, expanding access for developers and researchers to build applications on top of the model.

The company is also positioning pricing as a key part of its global AI strategy. Alibaba said developers can access Qwen3.8 through its API at a domestic price of 12 yuan ($1.67) per million input tokens and 36 yuan per million output tokens, with cached input tokens priced at 1.5 yuan.

International pricing for input and output tokens is set at just 40 percent and 24 percent of Anthropic's Opus 5 model, respectively.