Developer Gathin T. launched CleanYourTV, providing his parents' Xiaomi TV model to Claude Code, which connected over Wi-Fi, disabled 25 Xiaomi packages, and generated a new launcher with a cleaner interface. He later published the prompts and code, saying thousands of users tried similar modifications within two days.
On October 15, Xiaomi announced the full open-source release of MiDashengLM-Spatial, claiming it to be the industry's first open-source end-to-end spatial audio understanding large model. It retains basic audio understanding while adding spatial awareness and reasoning, and will be applied to scenarios such as smart homes and car cockpits.
Xiaomi internally released a promotion list, elevating 31-year-old LLM team lead Luo Fuli to Grade 22, setting the fastest promotion record in the company's history. Meanwhile, the MiMo team released and open-sourced the Xiaomi MiMo-V2.6 series, including Pro and Flash native all-modal models; Pro scored 46 on the Artificial Analysis Intelligence Index, ranking first among open-weight models.
Xiaomi's MiMo-V2.6 series, comprising three models, has gone live on OpenRouter, including a flagship with over 1 trillion parameters, an open-source MoE model, and a variant about 10 times faster than the flagship. All three support text, image, video, and audio inputs with a 1M context length.
Tech media NotebookCheck published a blog post based on Vulkan feature reports saying that Xiaomi's self-developed Xuanjie O3 chip supports BCn block-compressed texture formats (BC1 to BC7), which could pave the way for PC game compatibility. The Xiaomi 18 Fold and the Xiaomi Pad 9 Pro Max with the same chip support the formats, while MediaTek Dimensity and Google Tensor chips do not.
Post-00s security researcher Chaofan Shou disclosed on social media that he spent a five-figure sum in U.S. dollars to purchase a 6TB Fable model call dataset from a major domestic LLM proxy service provider. The dataset contains numerous core system credentials that could be used to take over internal systems of 19 technology companies, including Huawei, Xiaomi, NIO, and MiniMax, as well as seven national research institutions such as Zhangjiang Laboratory and the University of Science and Technology of China.
Xiaomi announced the release and open-sourcing of Xiaomi-CocktailASR-1, an industrial-grade target-speaker speech recognition LLM that uses an end-to-end architecture with a reference audio voiceprint prompt, achieves SOTA on multi-speaker benchmarks, and supports chain-of-thought reasoning.
<body><div>On July 16, Xiaomi launched Xiaomi-Robotics-1, an embodied foundation model for real-world mobile manipulation. Pretrained on 100,000 hours of real-world data w</div></body>
<body><div>Xiaomi Robotics reports 98% success in dual-side self-tapping nut fitting, nearing manual 99% pass rate, to be formally deployed. Robots also expand to assembly</div></body>
Xiaomi reorganizes its XiaoAi technical team in response to the AI large model trend, deeply integrating AI across all its businesses. This shift marks a transition from