Lei Jun posted a set of numbers on Weibo: Xiaomi will invest 60 billion RMB in AI over the next three years. At least 16 billion RMB in 2026 alone. Q1 R&D spend: 9 billion RMB, up 33.4% year over year. R&D headcount: 26,048. All historic highs.
Xiaomi is transforming from a phone company into an AI company. But look closely, and Xiaomi's AI map already seems fairly complete — the MiMo LLM, the miclaw phone Agent, the Agent Ecosystem Platform, 1 billion IoT devices, 746 million monthly active users. Everything that should be there is there.
Lei Jun clearly isn't satisfied. He's looking for a missing piece. Without it, the 60 billion RMB investment can't fully convert into business value.
That piece is MiMo SoloEngine.
1. Xiaomi's AI Map: Four Layers, Only the Last One Missing
Xiaomi Group President Lu Weibing laid out Xiaomi's three-layer AI architecture in detail on the 2026 Q1 earnings call. With the underlying infrastructure, it's actually four layers.
Layer 1: Infrastructure. Xiaomi holds roughly 220.6 billion RMB in cash reserves, 26,048 R&D staff, and 9 billion RMB in Q1 R&D spend. 60 billion RMB is the floor — actual investment will only be higher.
Layer 2: Foundation models. The MiMo model family is fully formed: V2.5-Pro (the flagship Agent model — trillion parameters, million-token context window), V2.5 (multimodal foundation), V2-Omni, V2-TTS, and the OneVL autonomous driving model. V2.5-Pro ranks first globally among open-source models on both Artificial Analysis's General Intelligence Index and Agent Index. Token efficiency is 40%–60% lower than Claude Opus 4.6 and GPT-5.4. MiMo Code, just released on June 11, pushes coding Agent capabilities to new heights.
Layer 3: AI application deployment. miclaw phone Agent has entered closed beta — China's first system-level AI Agent on mobile, with 50+ built-in system tools. The Agent Ecosystem Platform (dev.mi.com) has entered open beta. Miloco whole-home intelligence debuted at AWE2026. HyperOS connects 1.1 billion devices globally, with 746 million monthly active users.
Xiaomi's AI map looks complete. But CCID Consulting analyst Bai Runxuan pointed out a critical gap: the current Agent value chain shows a pattern of "hot at both ends, hollow in the middle" — upstream large models and chips are flush with capital, downstream scenario demand is strong, but the midstream lacks an engineering platform that can turn industry knowledge into reliable Agents.
Xiaomi's map perfectly confirms this diagnosis: the foundation models are there, the end-user devices are there, the ecosystem platform is there. What's missing is the bridge — a platform that lets "ordinary people" use these resources to build Agents.
2. The Gap Between 87% and 10%
At the AIGC2026 Summit, Amazon Web Services disclosed a data point: 87% of enterprises claim to have deployed AI at scale, but only 10% have actually gotten value from it.
The gap between these two numbers reveals a core contradiction: market demand for Agents is massive, but the ability to build Agents is locked inside the hands of developers.
Today there are only two paths to building an AI Agent. The first is low-code Workflow platforms like Dify and n8n. They offer visual canvases where users can drag and drop nodes to build AI applications. But the core logic is "preset paths" — using if/else conditions to control flow, with no support for true autonomous decision-making. Think of it like a subway map: every line and every stop is pre-planned, and trains can only run on fixed tracks.
The second path is code-based development frameworks like LangChain and CrewAI. They support true Agentic AI — Agents can make autonomous decisions and dynamically adjust strategies — but they require Python programming skills. A lawyer can't use LangChain. An accountant can't configure a ReAct Agent. An operations manager can't write Python.
This creates an obvious market gap: low-code platforms are easy to pick up, but don't support true autonomous decision-making; code frameworks support autonomous decision-making, but only developers can use them.
MiMo SoloEngine fills exactly this gap.
3. SoloEngine: The Bridge in Xiaomi's AI Ecosystem
SoloEngine is the first low-code Agentic AI development platform.
It packages the ReAct architecture, tool calling, the MCP protocol, Skills, and SubAgents all behind the scenes. Users open a browser, drag an Agent onto a canvas, wire up collaboration relationships, configure the tools they need, and click run. The backend automatically compiles everything into a dedicated Agentic AI system.
This system is not a Workflow — it doesn't follow a preset path. Each Agent runs a ReAct loop — think → act → observe → repeat — making real-time judgments based on the current situation. When it hits something unexpected, it adjusts its own strategy. When it finds a better approach, it switches paths on its own.
| Dimension | Dify/n8n | LangChain/CrewAI | SoloEngine |
|---|---|---|---|
| True Agentic AI support | ✗ Workflow only (preset paths) | ✓ ReAct / multi-Agent | ✓ ReAct / multi-Agent |
| Coding required | No | ✗ Must know Python | No |
| Visual orchestration | ✓ Full canvas experience | ✗ None | ✓ Full canvas experience |
| Domain experts can build independently | ✓ | ✗ | ✓ |
| Multi-Agent collaboration | ✗ | ✓ | ✓ |
SoloEngine also supports progressive disclosure — tools, Skills, and the MCP protocol load on demand, cutting Token consumption by over 85%. A unified adapter layer covers all major models — OpenAI, Anthropic, Ollama, MiMo, DeepSeek, Qwen, Zhipu, and more — with one interface for seamless switching. One-click packaging lets you bundle a finished Agent team into a complete product, ready for anyone to use.
4. SoloEngine Connects All of Xiaomi's Resources
MiMo's model capabilities, API costs reduced by 99%, 1 billion IoT devices, the Agent Ecosystem Platform, the miclaw phone Agent — these resources are connected by SoloEngine, forming an ecosystem moat that other platforms can't easily replicate.
The MiMo model provides SoloEngine's Agent "brain" — trillion parameters, million-token context, and the world's #1 open-source reasoning capability.
The 99% cheaper API pushes SoloEngine users' running costs to rock bottom — a 3-Agent collaboration team costs less than 300 RMB per month.
1 billion IoT devices let SoloEngine-orchestrated Agents directly control the physical world — from air conditioners to door locks, from cameras to robot vacuums.
The Agent Ecosystem Platform lets Agent teams built on SoloEngine be distributed and sold — going from "for my own use" to "for anyone to use."
The miclaw phone Agent extends SoloEngine's orchestration capabilities to mobile — users can direct Agent teams to execute tasks right from their phones.
While OpenAI is still locking AgentKit inside the GPT-5 ecosystem, Xiaomi has already combined MiMo with SoloEngine to bring the barrier for building Agents down to zero.
5. Infrastructure for the OPC Era
China's one-person limited liability companies have surpassed 16 million, accounting for 27.4% of all enterprises. 2026 is being called "Year One of OPC," with over 20 cities rolling out dedicated OPC support policies.
The core need of these "one-person companies" is using AI Agents to replace traditional teams — becoming a "one-person army." But LangChain requires programming skills, and Dify's Workflow doesn't support true autonomous decision-making.
SoloEngine is the infrastructure built for this era.
A lawyer uses SoloEngine to build a legal affairs Agent team — one person doing the work of three. An accountant uses SoloEngine to build a financial analysis Agent team — automatically generating reports and tax advice. An operations manager uses SoloEngine to build a marketing Agent team — one person running six online stores.
MiMo provides the Agent's "brain"; SoloEngine provides the Agent's "hands and feet." The final piece of the 60 billion RMB investment puzzle is finally in place.
SoloEngine's positioning: No Workflow. No orchestration code. Just Agents that get things done.
Top comments (0)