Alibaba has launched Qwen3.8-Max, a multimodal model with 2.4 trillion parameters, available through QwenCloud and Alibaba Cloud Model Studio at $2 per million input tokens and $6 per million output tokens. Built on the Qwen3.5 architecture with a sparse mixture-of-experts design and hybrid attention, it activates approximately 95 billion parameters per token. The model weights will be published on Hugging Face and ModelScope next week, marking the first Qwen-Max release with downloadable weights. Self-hosting requires storing and distributing full weights across multiple high-memory GPU nodes, making it practical mainly for large organizations and inference providers. Alibaba holds a 36% stake in Moonshot AI, whose Kimi K3 has 2.8 trillion parameters and topped Arena’s frontend coding leaderboard within hours of debut; demand overwhelmed Moonshot’s GPU capacity within 48 hours, pausing new subscriptions. In Alibaba’s tests, Qwen3.8-Max spent 16 days building a command-line app, took five days to reproduce a research paper’s results, and improved a chip design after about 500 iterations. The 16-day demo, building “oh-my-cli,” involved 265 commits, 127 pull requests, and 151 GitHub issues, with the public repository at qwen-code-dev-bot/oh-my-cli. Alibaba has not released enough information for outside researchers to