--- title: "Lei Jun officially announces the \"mysterious model\" that has sparked global developer discussions, Xiaomi's \"AI roadmap\" is gradually becoming clear" type: "News" locale: "en" url: "https://longbridge.com/en/news/279700638.md" description: "The \"mysterious model\" that has been widely discussed online has finally been claimed by Xiaomi! Xiaomi has launched the trillion-parameter model MiMo-V2-Pro, which ranks eighth globally and has significant cost advantages. Goldman Sachs released a research report supporting this, stating that Xiaomi has officially entered the phase of realizing AI achievements, solidifying its position as a \"physical AI leader.\" Although the tens of billions in R&D may pressure short-term profits, the ecological implementation will drive long-term value reshaping. Goldman Sachs maintains a \"buy\" rating with a bullish target of HKD 41" datetime: "2026-03-19T01:29:56.000Z" locales: - [zh-CN](https://longbridge.com/zh-CN/news/279700638.md) - [en](https://longbridge.com/en/news/279700638.md) - [zh-HK](https://longbridge.com/zh-HK/news/279700638.md) --- > Supported Languages: [简体中文](https://longbridge.com/zh-CN/news/279700638.md) | [繁體中文](https://longbridge.com/zh-HK/news/279700638.md) # Lei Jun officially announces the "mysterious model" that has sparked global developer discussions, Xiaomi's "AI roadmap" is gradually becoming clear Once speculated across the internet to be the "mysterious model" DeepSeek V4, it has ultimately been "claimed" by Xiaomi. This not only resolved the identity puzzle within the developer community but also provided a clearer reference for the capital market regarding how Xiaomi's AI investments will materialize. Recently, an AI model named Hunter Alpha was anonymously launched on the developer platform OpenRouter, attracting attention from the global developer community. Hunter Alpha went live on March 11 as an "invisible model," boasting over 1 trillion parameters and a 1 million token context window. Due to the parameter specifications closely matching the information circulating about DeepSeek V4, it sparked widespread speculation. Xiaomi ended the ongoing speculation with an official announcement. In the early hours of March 19, Xiaomi released a "triple update" for its MiMo large model series, including the flagship base model MiMo-V2-Pro, the multimodal agent model MiMo-V2-Omni, and the speech synthesis model MiMo-V2-TTS. Subsequently, Xiaomi founder Lei Jun stated on Weibo that Xiaomi had just released the trillion-parameter large model MiMo-V2-Pro and disclosed that Xiaomi's "actual progress in AI may be much faster than what everyone sees," with AI research and capital investment expected to exceed 16 billion this year. **In response, Goldman Sachs pointed out in a research report released on March 19 that the concentrated release of these three flagship models marks Xiaomi's transition from the AI research and development investment phase to the results realization phase, with its market positioning as a "physical AI leader" gradually gaining substantial support. They maintained a "buy" rating on Xiaomi, with a 12-month target price of HKD 41, implying about a 14% upside from the current stock price.** ## Xiaomi's Three Models Make a Grand Debut **Xiaomi MiMo-V2-Pro is designed specifically for high-intensity agent work scenarios in the real world.** **It has a total parameter count of over 1T (42B active parameters) and supports a super long context length of 1M.** In terms of underlying architecture design, it inherits a hybrid attention mechanism and significantly increases the hybrid ratio from 5:1 to 7:1, balancing ultra-large scale with extremely high inference efficiency. MiMo-V2-Omni is positioned as Xiaomi's multimodal foundational model, integrating multimodal understanding capabilities of images, videos, and audio with powerful agent capabilities. According to the Goldman Sachs report, **this model reaches or exceeds the levels of Gemini 3 Pro, Claude Opus 4.6, Gemini 3, and GPT-5.2 in several core metrics, including audio understanding, image understanding, video understanding, and agent capabilities.** MiMo-V2-TTS is aimed at the era of voice intelligent agents, providing highly controllable multi-granularity style control, natural rhythm reproduction, and singing capabilities. Goldman Sachs pointed out that its next goal includes expanding language coverage beyond Chinese and English, and deeply integrating with the multimodal understanding capabilities of MiMo-V2-Omni, enabling agents to describe the real world with a voice close to human expressiveness. The three models have been integrated into the miclaw intelligent agent system on WPS Office, Xiaomi phones, and computers, as well as the Xiaomi browser. ## Flagship Model Performance: Eighth in the World, Significant Cost Advantage, Deeply Optimized for Agent Scenarios MiMo-V2-Pro is the core of this release. This model has over 1 trillion total parameters, 42 billion active parameters, and a context window of 1 million tokens, ranking eighth globally and second among Chinese models on the Artificial Analysis Intelligence Index, surpassing xAI Grok, and only behind Gemini 3.1 Pro Preview, GPT-5.4, GPT-5.3 Codex, Claude Opus 4.6, Claude Sonnet 4.6, and GLM-5. **Goldman Sachs pointed out that cost efficiency is another core competitiveness of MiMo-V2-Pro.** According to Artificial Analysis data, the cost of running this model to complete the Intelligence Index test is $348, which is 36% lower than GLM-5, which ranks higher on the same list, and 90% lower than Claude Sonnet 4.6; compared to Claude Opus 4.6 and Claude Sonnet 4.6, the token usage cost of MiMo-V2-Pro is up to 80% lower. In terms of training efficiency, Xiaomi's recently launched ARL-Tangram system has been deployed to support the training of the MiMo series models, achieving an average action completion time improvement of 4.3 times, a reinforcement learning training speed increase of up to 1.5 times, and external resource savings of up to 71%. **It is worth mentioning that MiMo-V2-Pro is deeply optimized for Agent scenarios.** MiMo-V2-Pro conducts SFT & RL for complex and diverse Agent Scaffolds, possessing stronger tool invocation and multi-step reasoning capabilities. On the OpenClaw standard evaluation lists PinchBench and ClawEval, MiMo-V2-Pro's performance is among the world's top. Additionally, with a super long context window of 1M, MiMo-V2-Pro can comfortably support high-intensity real Claw complex application flows ## Clarifying the AI Roadmap: System Layout from Models to Ecosystems Goldman Sachs reports that this release is not an isolated event, but rather a systematic advancement of Xiaomi's acceleration in transforming AI research and development investments into actual results. **Prior to this, Xiaomi released the visual-language-action model Xiaomi-Robotics-0 for robotic reasoning and real-time execution in February this year, followed by the AI agent system miclaw in March, and on March 19, launched an upgraded assisted driving system HAD equipped with the XLA cognitive model, which integrates Xiaomi's self-developed embodied foundational model MiMo-Embodied.** Goldman Sachs outlined clear iterative directions for the three models: the next goal for MiMo-V2-Pro is to tackle high-complexity reasoning and long-cycle task planning; the goal for MiMo-V2-Omni is to achieve continuous intention planning across hours and even days, real-time streaming perception, and executing actions through robots and hand movements; MiMo-V2-TTS will expand to multiple languages and deepen integration with Omni. Goldman Sachs believes that with its leading multimodal AI capabilities and the rich application scenarios of intelligent agents within the "people-vehicle-home" ecosystem, Xiaomi has the potential to capture a vast market space in the global AI model industry, while also aiming to create high-premium, differentiated consumer-grade AI terminals. ## Increased Investment and Valuation Logic: Short-term Profit Pressure, Long-term Value Reassessment Goldman Sachs expects Xiaomi's R&D expenditure to reach RMB 40 billion by 2026, up from the estimated RMB 32.2 billion in 2025, with the continued increase in investment likely to weigh on recent profits. Financial data shows that Goldman Sachs predicts Xiaomi's net profit (excluding special items) for 2026 to be approximately RMB 27.9 billion, lower than the estimated RMB 39.5 billion for 2025, corresponding to a price-to-earnings ratio of about 27.4 times for 2026. Nevertheless, Goldman Sachs believes that continuous delivery of results should prompt the market to reprice Xiaomi as a physical AI leader with self-developed AI, operating systems, and chip capabilities, rather than merely valuing it based on recent price-to-earnings ratios. Goldman Sachs maintains a buy rating with a target price of HKD 41, based on a sum-of-the-parts valuation method, including a 16 times target 12-month forward EV/NOPAT for Xiaomi's core business, a DCF valuation (USD 45 billion) for Xiaomi's electric vehicle business, and applying a 10% holding discount ### Related Stocks - [XIAOMI-W (01810.HK)](https://longbridge.com/en/quote/01810.HK.md) - [Xiaomi Corporation (XIACY.US)](https://longbridge.com/en/quote/XIACY.US.md) ## Related News & Research - [Poco’s first Pro Max phone earns the name with an 8,500mAh battery](https://longbridge.com/en/news/279438433.md) - [Xiaomi bought back 2.3 mln type B shares for HKD76.9 mln on Mar 12, HKEX filing shows](https://longbridge.com/en/news/278864067.md) - [First Verifiable AI Agents Go Live on Mainnet: Cysic AI Launches Autonomous Agent Swarms with Cryptographic Proof](https://longbridge.com/en/news/279479162.md) - [Alset AI Converts Debt to Equity to Bolster Balance Sheet](https://longbridge.com/en/news/279093528.md) - [12 AI stocks that will skyrocket](https://longbridge.com/en/news/279274043.md)