Alibaba released the new AI model Qwen-3.8-Max-0902, which immediately topped the Code Arena WebDev rankings. At the same time, in terms of the ratio of features to cost, it is clearly ahead of the more expensive solutions of its competitors, including the Anthropic model.
Qwen-3.8-Max-0902 has 2.4 trillion parameters and costs $2 per million input tokens and $6 per million output tokens. The cost of an explicit cache hit is $0.17 and the cost of an implicit cache is $0.25 per million tokens. In Code Arena WebDev, the model achieved a score of 1629, second only to Claude Opus 5 (Max) with a score of 1688, but the latter has a value of $20 per million input tokens. Qwen-3.8-Max-0902 also ranked highly in the Data & Analytics and Expenditure Products categories, ranked second in the Branding & Marketing, Games & Simulations category, and ranked third in the Content Creation Tools and Reference-Based Design categories.
The new model launches approximately two weeks after the launch of Qwen-3.8 27B, an open model with 27 billion parameters that demonstrates comparable programming capabilities to Opus 4.5 and is capable of running on a single MacBook. Alibaba’s rapid progress has led to speculation about the sublimation of others’ models, but there is no evidence of this. Notably, the company itself recently banned employees from using Claude Code, saying the tool was high-risk.