Alibaba has launched Qwen3.8-Max, its largest AI model yet with 2.4 trillion total parameters, and says it will release the open weights next week. The announcement lands a week after Moonshot AI’s free Kimi K3, intensifying the race among Chinese labs to ship frontier-scale open models.
The flagship model uses a mixture-of-experts architecture that activates only about 95B parameters per request, keeping inference costs manageable at trillion-parameter scale. It handles text, images and video, and supports a context window of up to one million tokens. Alibaba says Qwen3.8-Max completed a 16-day software engineering project, and the API is already live on its Qwen platform and integrated into the company’s new Qwen Office agent.
Alongside the flagship, Alibaba is open-sourcing Qwen3.8-27B, a smaller model designed for easier deployment. The company positions the releases as a coding and professional office focus, giving developers a choice between its largest model and a lightweight option.
Qwen3.8-Max quickly became the highest-ranked Chinese text model on Arena.AI after its debut, and climbed to second place globally in multimodal testing behind a Claude variant. Alibaba shares jumped 7% in Hong Kong trading following the launch.
The release comes as Chinese AI labs compete on both capability and cost. DeepSeek’s V4-Flash, priced at $0.14 per million input tokens, undercuts Western rivals by a wide margin on benchmark economics, and Moonshot’s Kimi K3 is the largest free model released to date. Open weights and aggressive pricing give developers a route to frontier intelligence that US labs have largely kept behind APIs.
