Alibaba open-sources Qwen3.8-Max with 2.4T params and revenue-sharing
Qwen3.8-Max's 95B active parameters and 1M context hits Hugging Face this week
Alibaba's Tongyi Qwen team officially released Qwen3.8-Max on August 3, and the open-source weights are scheduled for public download via Hugging Face and ModelScope during the week of August 10, 2026. The model is a 2.4-trillion-parameter Sparse Mixture of Experts (MoE) architecture that activates only 95 billion parameters per inference, making it significantly more compute-efficient than dense models of comparable scale. It also features a 1-million-token context window—enough to process entire codebases or lengthy documents in a single pass—and native multimodal support for text, image, audio, and video inputs.
Beyond the open weights, Alibaba has reportedly adopted a revenue-sharing scheme for large commercial users of Qwen3.8-Max. This monetization approach, which follows Moonshot AI's Kimi K3 License, marks a notable shift for open-weight AI: companies can download and fine-tune the model, but if they generate substantial revenue using it, they share a portion with Alibaba. This balances open accessibility with commercial sustainability, setting a precedent that could reshape how frontier AI models are distributed. For developers, Qwen3.8-Max offers enterprise-grade performance without API lock-in, while large businesses must now factor revenue-sharing into their cost models.
- Qwen3.8-Max is a 2.4T-parameter sparse MoE model with only 95B active parameters per inference
- Includes a 1M-token context window and native multimodal support for text, image, audio, and video
- Open-source weights arrive August 10, 2026, with revenue-sharing for large commercial users like Moonshot AI's Kimi K3 License
Why It Matters
A 2.4T-param open model with revenue-sharing could reset economics for enterprise AI and open-weight licensing.