In 2026, when artificial intelligence technology is rapidly iterating, technology giant Xiaomi quietly launched three self-developed large-scale language models early this morning: MiMo-V2-Pro, MiMo-V2-Omni and MiMo-V2-TTS, and also officially launched the Xiaomi MiMo Claw intelligent assistant service. This series of actions not only demonstrates Xiaomi’s profound technological accumulation in the field of AI base models, but also marks its determination to officially enter the AI Agent application ecosystem. From flagship base models to full-modal understanding capabilities, to highly immersive speech synthesis technology, Xiaomi is trying to build an AI ecosystem covering the entire link of “understanding – reasoning – expression.”
Xiaomi announces three self-developed MiMo-V2 series models and launches MiMo Claw smart assistant
MiMo-V2-Pro: The flagship base model built for the Agent era
Xiaomi MiMo-V2-Pro is the highlight product released by Xiaomi this time. It is specially designed for high-intensity Agent work scenarios in the real world. The model has over 1T (one trillion) total parameters, uses a hybrid attention architecture with 42B (four and two billion) activation parameters, and supports ultra-long context lengths of up to 1M (one million tokens). This specification enables it to handle complex tasks such as large-scale code base analysis and long document processing.
Introducing MiMo-V2-Pro & Omni & TTShttps://t.co/qdufHaBJWY pic.twitter.com/m5YFmV37WB
— Xiaomi MiMo (@XiaomiMiMo) March 18, 2026
In terms of performance, MiMo-V2-Pro ranks eighth in the world and second in China on the Artificial Analysis rankings. What’s more noteworthy is that in the actual testing of agent frameworks such as OpenClaw and Claude Code, this model can complete complex workflow orchestration, long-range planning and precise tool invocation without manual intervention. It is said that the overall somatosensory experience has surpassed Claude Sonnet 4.6 and is close to the level of Opus 4.6.
However, the real killer feature of the MiMo-V2-Pro is its extremely competitive pricing strategy. Compared with the high cost of use of Claude Opus 4.6, MiMo-V2-Pro’s API pricing is only one-fifth of it. Specifically, the input fee within 256K contexts is approximately NT$ 31.9 (USD $1) per million tokens, and the output fee is approximately NT$ 95.7 (USD $3); while the input fee within 1M context is approximately NT$ 63.8 (USD $2), and the output fee is approximately NT$ 191.4 (USD $6).
In addition, MiMo-V2-Pro has now fully integrated into China’s popular Kingsoft WebOffice ecosystem, natively supporting the four mainstream document formats of Word, Excel, PPT, and PDF, seamlessly covering more than 95% of daily document types. WPS Lingxi has also been connected to this model, and users can directly ask questions or assign tasks to Lingxi Claw.
MiMo-V2-Omni: a new benchmark for full-modal understanding
MiMo-V2-Omni is an all-modal base model launched by Xiaomi for the Agent era. It is designed for complex multi-modal interaction and execution scenarios in the real world. This model can be seamlessly connected to various Agent frameworks, achieving a leap from understanding to manipulation, and significantly lowering the threshold for the implementation of full-modal Agents.
In terms of audio understanding, MiMo-V2-Omni supports everything from environmental sound classification, multi-speaker separation, audio-visual joint reasoning, to deep understanding of more than 10 hours of continuous long audio. Its overall performance surpasses Gemini 3 Pro and is one of the strongest audio understanding base models currently available.
In terms of image understanding, MiMo-V2-Omni demonstrates powerful multi-disciplinary visual reasoning and complex chart analysis capabilities, claiming to surpass Claude Opus 4.6 and approaching the level of top closed-source models such as Gemini 3 Pro. In terms of video understanding, the model supports native audio and video joint input to achieve true multi-modal video understanding, and has strong situational awareness and future reasoning capabilities.
With these capabilities, MiMo-V2-Omni is able to understand complex environments across modalities, autonomously formulate and execute plans, revise strategies in real time when encountering anomalies, and ultimately deliver complete results end-to-end. The model is now open to API services and supports 256K context length. The input fee is approximately NT$ 12.8 (USD $0.4)/million tokens, and the output fee is approximately NT$ 63.8 (USD $2)/million tokens.
MiMo-V2-TTS: Highly controllable speech synthesis large model
Xiaomi MiMo-V2-TTS is a large speech synthesis model independently developed by Xiaomi. It is based on the self-developed Audio Tokenizer and multi-codebook speech-text joint modeling architecture. After large-scale pre-training and multi-dimensional reinforcement learning of hundreds of millions of hours of speech data, the model achieves highly controllable, multi-granular speech style control.
The core advantage of MiMo-V2-TTS lies in its rich multi-expression capabilities. Users can set the overall voice tone through natural language instructions, and at the same time perform fine-grained emotional adjustment on local segments within the sentence, achieving natural transitions between tone transitions and emotional gradients in the same sentence. The model supports the natural pronunciation of multiple dialects, including Northeastern dialect, Sichuan dialect, Henan dialect, Cantonese, Taiwanese accent, etc. It can perform role-playing stylized interpretations and achieve high-quality singing synthesis – allowing the same model to speak, act, and sing.
MiMo Claw Intelligent Assistant: One-click deployment of AI assistant
Launched simultaneously with the three large models, there are also Xiaomi MiMo Claw Intelligent assistant service. Users can experience this “Openclaw” assistant for free through the MiMo Studio official website, and each experience lasts 30 minutes. According to the official introduction, Xiaomi MiMo Claw can help users complete various tasks such as document generation, news acquisition, content creation, development efficiency improvement, and data analysis. This tool uses a regular conversation form and comes with a file system. Users can capture pictures and news from the website and store them in files. After exiting the experience, relevant data will be destroyed to protect user privacy.
The core highlights of MiMo Claw include: equipped with the latest flagship base model of MiMo-V2-Pro and MiMo-V2-Flash-Omni multi-modal understanding model; one-click deployment of OpenClaw, zero-cost experience; built-in diverse skills to easily complete complex tasks; integrated Kingsoft WebOffice online document preview, supporting the four mainstream formats of Word, Excel, PPT, and PDF. It is currently unknown whether an international version will be launched.
point of view
Xiaomi launched three large models in one go and launched the MiMo Claw intelligent assistant, demonstrating its ambition and strength in the field of AI. From a technical perspective, the MiMo-V2 series has reached or approached the top international level in multiple benchmark tests. Especially in the optimization of Agent scenarios, Xiaomi has chosen a path of deep integration with frameworks such as OpenClaw and Claude Code. This strategy helps to quickly establish a developer ecosystem.
What deserves more attention is its pricing strategy. In the context of the current generally high price of large language model APIs, MiMo-V2-Pro provides close performance at only one-fifth the price of Claude Opus 4.6. This “high cost performance” route is in line with Xiaomi’s past strategy in the hardware field. However, whether the price advantage can be converted into market share depends on its stability and developer experience in actual applications.
Overall, this series of actions by Xiaomi marks its official entry into the competition of AI base models. In this dual battle between technology and ecology, whether Xiaomi can stand out with its “high cost performance + China’s local ecological integration” strategy deserves continued attention.
Source: KOCPC Chinese