Alibaba, ByteDance and Xiaomi push ahead on China AI compute
The day’s China AI compute news centered on new model releases, enterprise agent tools, and in-house chip and hardware demos from major tech firms.
• Alibaba said Qwen3.8-Flash-Next and Qwen3.8-Flash-Next-FP8 will be open-sourced at 23:00 Beijing time on Aug. 26; Qwen3.8-Flash-Next is a multimodal MoE model built on the next-generation Qwen4 architecture.
• Alibaba Cloud said its Token Plan is now integrated into the Qwen app and PC client, letting users bind an API key and use subscribed quota with models including Qwen3.8-Max.
• ByteDance officially launched Doubao for Work, an agent product deeply integrated with Feishu for enterprise workflows, and is offering a free 30-day subscription benefit to new users who download and log in or upgrade.
• Tencent Cloud and KargoBot formed a strategic partnership to deepen cooperation on AI infrastructure for training, inference, data processing and continuous model iteration for autonomous-driving freight operations.
• Xiaomi showed the Xring O100 prototype and AI Cube hardware, describing coordinated dual-chip computing between Xring O3 and O100, an active air-cooling system and support for built-in Xiaomi MiMo on-device models.
• Xiaomi said it has released three in-house chips, including Xring O3, Xring O100 and D100, as part of its push into self-developed silicon and on-device AI.