TLDR: Xiaomi has launched MiDashengLM-7B, an open-source AI voice model designed to enhance its smart home and automotive products. This new model offers offline functionality, faster responses, and advanced multi-sound recognition, aiming to create a seamless ‘Human x Car x Home’ intelligent ecosystem.
Beijing, China – August 2025 – Xiaomi has officially launched its groundbreaking open-source AI voice model, MiDashengLM-7B, marking a significant leap in the company’s ambition to create a unified and intelligent ‘Human x Car x Home’ ecosystem. Unveiled in early August 2025, this new voice assistant is specifically optimized for seamless integration across Xiaomi’s smart home devices and automotive systems, promising enhanced user experiences with its advanced capabilities.
The MiDashengLM-7B model distinguishes itself with several key features designed to overcome the limitations of traditional voice assistants. A standout innovation is its robust offline functionality, allowing the AI to operate without constant reliance on cloud servers, thereby ensuring faster responses and uninterrupted service even in areas with spotty internet connectivity. This also minimizes additional costs associated with cloud access.
Built upon a combination of Xiaomi’s proprietary Dasheng audio encoder and Alibaba’s open-source Qwen2.5-Omni-7B decoder, MiDashengLM-7B is capable of interpreting not only human speech but also a wide array of ambient and environmental sounds. This multi-sound recognition enables gesture-based controls, such as responding to claps or snaps, and provides real-time audio surveillance, sending alerts upon detecting unusual noises in a room. The model also offers pronunciation feedback for language learning, making it a versatile tool for various applications.
Xiaomi states that MiDashengLM-7B comes equipped with over 30 AI-powered features. In smart homes, it functions as a comprehensive safety guard and responds to complex voice commands, while in vehicles, it facilitates voice-activated controls. A particularly innovative feature is ‘underwater voice wake-up,’ allowing users to interact with devices in wet environments without physical contact. Furthermore, the model boasts low-latency performance and can handle twice as many tasks simultaneously compared to previous systems, significantly boosting efficiency.
The release of MiDashengLM-7B as an open-source model under the Apache 2.0 license underscores Xiaomi’s strategic commitment to fostering innovation and attracting developers. This move aligns with a growing industry trend where open-source AI models are favored for their cost-effectiveness and ability to capture developer mindshare, providing a competitive edge against proprietary alternatives. This strategy is a culmination of years of systematic AI investment by Xiaomi, which established dedicated artificial intelligence, big data, and cloud computing departments in February 2019. Xiaomi CEO Lei Jun has previously emphasized the critical importance of technology, stating, ‘technology is a matter of life and death for Xiaomi.’ The company’s long-standing partnership with Microsoft since 2018 for Azure cloud integration has also laid a strong foundation for its extensive IoT ecosystem, which by 2017 already included 85 million connected devices.
Also Read:
- DeepSeek Unveils V3.1 AI Model with 685 Billion Parameters on Hugging Face
- xAI Unveils Grok 4 and Announces Free Access, Reshaping AI Landscape
The global voice assistant application market is experiencing rapid growth, valued at USD 4.74 Billion in 2024 and projected to reach USD 34.26 Billion by 2032, with a Compound Annual Growth Rate (CAGR) of 30.94% from 2026 to 2032. Xiaomi’s MiDashengLM-7B is poised to capture a significant share of this expanding market, reinforcing its position as a key player in the global AI landscape. The model is already live in Xiaomi’s smart TVs and car systems across China, demonstrating its immediate applicability and the company’s readiness to deploy advanced AI solutions.
