--- title: "Xiaomi launches a new AI model claiming its performance ranks among the top two global open-source large models" type: "News" locale: "zh-HK" url: "https://longbridge.com/zh-HK/news/269956362.md" description: "Xiaomi President Lu Weibing announced the launch of the self-developed AI large model MiMo-V2-Flash, which ranks among the top two global open-source large models in performance. This model supports deep thinking and internet search, providing an intelligent user experience. AI head Luo Fuli introduced its three major features: super strong code invocation capability, extreme reasoning efficiency, and a new post-training paradigm. MiMo-V2-Flash performs excellently in code and Agent evaluation benchmarks, with a reduced number of parameters but performance surpassing other models" datetime: "2025-12-17T04:40:55.000Z" locales: - [zh-CN](https://longbridge.com/zh-CN/news/269956362.md) - [en](https://longbridge.com/en/news/269956362.md) - [zh-HK](https://longbridge.com/zh-HK/news/269956362.md) --- > 支持的語言: [简体中文](https://longbridge.com/zh-CN/news/269956362.md) | [English](https://longbridge.com/en/news/269956362.md) # Xiaomi launches a new AI model claiming its performance ranks among the top two global open-source large models Xiaomi (1810) President Lu Weibing announced at the 2025 Xiaomi People Vehicle Home Ecosystem Partner Conference that Xiaomi's self-developed AI large model Xiaomi MiMo-V2-Flash has officially been open-sourced and launched, referring to it as a new language foundation for entering the Agent era. Xiaomi MiMo is the first inference large model independently developed by Xiaomi. The company stated that MiMo-V2-Flash not only supports deep thinking but also has internet search capabilities, enabling natural and smooth conversations with users while playing an important role in scenarios requiring real-time data, the latest updates, or data verification, providing users with a more convenient and intelligent experience. According to the timeline announced by Lu Weibing for Xiaomi's self-developed MiMo series, Xiaomi has already launched the inference large model MiMo-7B, the visual inference large model MiMo-VL, the native end-to-end audio generation model MiMo-Audio, the edge-side visual language large model MiMo-VL-Miloco, and the embodied large model MiMo-Embodied. ## AI Genius Luo Fuli Makes Her First Public Appearance The head of the large model, Luo Fuli, made her first public appearance at the conference. This AI genius and former DeepSeek researcher introduced the newly open-sourced Xiaomi MiMo-V2-Flash, highlighting three major features. She explained that first, MiMo-V2-Flash has strong code and tool invocation capabilities, using the efficient "language" of Agent communication to make information exchange smoother; second, it is meticulously designed around extreme inference efficiency, which is key to achieving "high bandwidth" Agent collaboration, providing strong support for the collaborative processing of complex tasks; third, it adopts a new post-training paradigm, continuously optimizing model performance through efficient and stable expansion reinforcement learning training. Luo Fuli further stated that Xiaomi MiMo-V2-Flash contains strong foundational potential. Despite having a total parameter count of only 309B, it ranks among the top 2 global open-source models in terms of code and Agent evaluation benchmarks, even surpassing DeepSeek V3.2 and K2-Thinking in most evaluation benchmarks, while reducing the parameter count by 1/2 to 2/3, truly achieving a big impact with a small model. Additionally, it was introduced that MiMo-V2-Flash also possesses extreme inference efficiency, standing out in the speed and cost quadrant of the world's top models with its low cost and high speed, demonstrating strong competitiveness. ## One of the Highest Cost-Performance High-Performance Models Available Luo Fuli also clearly stated on social media, "MiMo-V2-Flash is now online. This is just the second step on our AGI roadmap." This statement highlights Xiaomi's long-term planning and technological ambition in the AI field. From a market impact perspective, the entry of MiMo-V2-Flash may disrupt the existing competitive landscape of open-source AI models. Its officially announced extremely low cost of $0.1 per million input tokens and $0.3 per output token, combined with an inference speed of up to 150 tokens/second, makes it one of the highest cost-performance high-performance models available Currently, the model is temporarily available for free on the API platform and has released the basic weights under the MIT open-source license on Hugging Face ### 相關股票 - [Xiaomi HK SDR 2to1 (HXXD.SG)](https://longbridge.com/zh-HK/quote/HXXD.SG.md) - [Xiaomi Corporation (XIACY.US)](https://longbridge.com/zh-HK/quote/XIACY.US.md) - [XIAOMI-W (01810.HK)](https://longbridge.com/zh-HK/quote/01810.HK.md) ## 相關資訊與研究 - [Guosheng Securities Sticks to Their Buy Rating for Xiaomi (XIACF)](https://longbridge.com/zh-HK/news/281478308.md) - [Xiaomi Streamlines Shareholdings and Renews VIE Contracts for Mainland Units](https://longbridge.com/zh-HK/news/281559343.md) - [CoreWeave Delivers Leading Inference Performance in MLPerf® Benchmark | CRWV Stock News](https://longbridge.com/zh-HK/news/281390933.md) - [Insig AI Plans Growth Drive and Eyes Nasdaq Dual Listing](https://longbridge.com/zh-HK/news/281311983.md) - [12:01 ETd-Matrix Boosts Rack-scale AI Capabilities With Acquisition of GigaIO Data Center Business](https://longbridge.com/zh-HK/news/281552365.md)