Xiaomi Corp. has released MiDashengLM-7B, an open-source voice model designed to run its electric vehicles and smart-home appliances. The seven-billion-parameter system, published under an Apache 2.0 license, builds on Xiaomi’s in-house speech technology and Alibaba’s Qwen2.5-Omni-7B. The company said the software is already deployed across its car dashboards and Internet-of-Things ecosystem, offering wake-word activation, natural-language commands and on-device inference. By opening the code and training data, Xiaomi aims to attract outside developers and reduce reliance on proprietary U.S. platforms. Internal benchmarks cited by the Beijing-based firm show MiDashengLM-7B matching or surpassing commercial rivals on Chinese and English voice tasks while running on consumer-grade GPUs. The move highlights intensifying competition among Chinese tech companies to push AI capabilities onto edge devices. Hours after Xiaomi’s announcement, Tencent’s HunYuan lab unveiled four compact language models ranging from 0.5 billion to 7 billion parameters, each capable of 256,000-token contexts on a single GPU. Analysts say the rapid stream of open-source releases could accelerate adoption of domestic AI tools across the nation’s auto and appliance sectors.