Lei Jun's Next Move After LLM, Reshapes the Agent Landscape

Dufrence — Wed, 03 Jun 2026 15:56:18 +0000

In the early hours of May 27, Xiaomi's tech team dropped an announcement: the MiMo-V2.5 API lineup would see permanent price cuts of up to 99%, with tiered pricing by context window eliminated entirely. Lei Jun reshared the news.

The price cut itself wasn't surprising—DeepSeek had done the same thing two weeks prior, and the market barely flinched. What really caught the industry's attention was a separate piece of news: on June 3, Xiaomi released the latest version of its open-source project, SoloEngine. The first low-code Agentic AI development platform.

I. Xiaomi's AI Year in Review: From 7B to a Trillion-Parameter Powerhouse

Xiaomi's pace in AI has exceeded most expectations.

In April 2025, MiMo-7B went open source. A 7B-parameter model outperforming OpenAI o1-mini on mathematical reasoning and coding benchmarks initially raised eyebrows. It wasn't until Artificial Analysis verified the results on its independent leaderboard that the skepticism subsided.

By late 2025, Lei Jun had recruited Luo Fuli from DeepSeek. Born after 1995, she was a core developer of DeepSeek-V2, dubbed the "AI prodigy" within the industry. She assembled a hundred-person team at Xiaomi with an average age of 25, over sixty percent of whom graduated from Tsinghua and Peking University. Achieving top-tier parity among global open-source models with a core team of just 100—that's exceptionally rare in the industry.

At the Spring 2026 launch event, Lei Jun unveiled three proprietary LLMs in one go—MiMo-V2-Pro, V2-Omni, and V2-TTS—while pledging 60 billion yuan in AI investment over the next three years. Around the same time, miclaw, the phone-based Agent, entered closed beta. It was China's first system-level AI agent on a mobile device, packing over 50 system tools capable of autonomously decomposing tasks, invoking system functions, and coordinating with Mijia IoT devices.

In April, the MiMo-V2.5 series went open source under the MIT license. On launch day, it was already adapted to seven domestic chip platforms—Alibaba's T-Head, AMD, Baidu's Kunlun, Enflame, Muxi, Tianshu Zhixin, and Amazon Web Services. A concurrent 100-trillion-Token incentive program distributed 100 trillion tokens free of charge within 30 days. Xiaomi's Agent ecosystem platform also entered public beta during this period.

In May, MiMo-V2.5-Pro scored 1581 on the GDPVal-AA benchmark, surpassing DeepSeek-V4-Pro and GLM-5.1 to claim the top spot among global open-source models. Hermes Agent topped the global usage charts on OpenRouter, hitting 291 billion tokens per day in call volume. The OneVL autonomous driving model went open source. Then came the 99% permanent price cut.

In internal testing, MiMo-V2.5-Pro completed the SysY compiler project from Peking University's compiler theory course in 4.3 hours—a project that typically takes Peking University undergraduates weeks to finish—scoring a perfect 233/233 with 672 tool invocations along the way.

From a small 7B model to a trillion-parameter flagship, from a follower to a frontrunner. Xiaomi pulled it off in under a year.

But no matter how powerful a foundation model is, it's essentially just a "brain." What Lei Jun wants to do next is give that brain a pair of hands.

II. How SoloEngine Is Reshaping the Agent Landscape

To understand why SoloEngine matters, you first need to grasp a key distinction: the difference between Agentic AI and workflows.

Agentic AI has already gained traction among developers. Claude Code can handle requirement decomposition, code writing, testing, and deployment from a single terminal prompt. ByteDance's Trae, in its SOLO mode, lets AI develop autonomously while developers only need to review and approve. But these tools are developer-only—building AI applications for non-programmers has traditionally come down to two approaches.

The first is low-code workflow platforms like Dify and n8n. These offer visual canvases where users can drag and drop nodes, connect data flows, and quickly assemble an AI application. But their core logic relies on "preset paths"—you lay out step A to step B to step C on the canvas and use if/else conditions to control branching. The entire process is like a subway map: every line and every stop is planned in advance, and the train can only run on fixed tracks. Hit something unplanned, and the flow breaks.

The second is code-based development frameworks like LangChain and CrewAI. These require Python programming skills, letting users define an AI Agent's role, tools, and collaboration logic through code. They do support genuine Agentic AI—agents can make autonomous decisions and dynamically adjust strategies—but the barrier to entry is steep. A lawyer won't use LangChain. An accountant can't configure a ReAct Agent. A marketing manager doesn't write Python.

This creates an obvious market gap: low-code platforms are easy to use but don't support true autonomous decision-making; code frameworks support autonomous decision-making but are only accessible to programmers.

Manus and the viral OpenClaw from early 2026 tried the generalist route—capable of doing everything, but excelling at nothing. General-purpose Agents struggle to meet commercial requirements.

SoloEngine fills precisely this gap.

It takes the vertical approach—empowering domain experts in every industry to build precise, reusable Agentic AI tools for their specific tasks, rather than trying to be a jack-of-all-trades assistant.

SoloEngine uses an Agentic AI architecture where each Agent runs a "think → act → observe → repeat" loop. This means Agents don't follow preset paths; they assess the situation in real time and make decisions on the fly. When something unexpected happens, they adjust their strategy on their own. When a better approach emerges, they switch routes proactively. The entire process requires no if/else conditions from the user and no pre-planning of every possible path.

Using SoloEngine is fundamentally different from anything else. Users open a browser, drag Agents onto a canvas, connect collaboration relationships, configure the tools they need, and hit run. The backend automatically compiles the visual design into an executable Agentic AI system—one that can plan tasks, execute operations, and deliver real-time feedback, while the user only needs to review and confirm.

No lines of code. No if/else logic to configure.

Take a concrete scenario: a lawyer drags a "Contract Review Agent" onto the canvas, adds a "Legal Statute Search Agent" and a "Risk Flagging Agent," connects their collaboration relationships, and hits run. Thirty minutes later, a contract review report with 37 flagged risk points is automatically generated. The entire process is zero-code.

Here's how SoloEngine stacks up against the mainstream options:

	Dify/n8n	LangChain/CrewAI	SoloEngine
True Agentic AI support	✗ Preset-path workflows only	✓ ReAct / multi-Agent	✓ ReAct / multi-Agent
Programming required	No	✗ Must know Python	No
Visual orchestration	✓ Full canvas experience	✗ None	✓ Full canvas experience
Can domain experts build independently	✓	✗	✓
Multi-Agent collaboration	✗	✓	✓

Progressive disclosure—tools, Skills, and MCP protocols load on demand, so Agents only invoke the tools they actually need, cutting token consumption by over 85% in complex tasks. Unified adaptation layer—covering all major models including OpenAI, Anthropic, Ollama, MIMO, DeepSeek, Tongyi Qianwen, and Zhipu, with a single interface for seamless switching and no lock-in to any single vendor. One-click packaging—assembled Agent teams can be packaged into complete products for anyone to use directly.

III. Why Xiaomi Is Entering the Agent Space

Xiaomi already has the MiMo foundation model. Why invest resources in building a development platform?

China's enterprise AI agent market is projected to surpass 43 billion yuan in 2026 (IDC data). Yet at the AIGC2026 Summit, Amazon Web Services disclosed a striking statistic: 87% of enterprises claim to have deployed AI at scale, but only 10% have actually extracted real value from it. The gap between these two figures reveals a core contradiction—demand for Agents is enormous, but the ability to build them remains locked in the hands of programmers.

Bai Runxuan, an analyst at CCID Consulting, put it vividly: the current agent industry chain exhibits a "hot at both ends, hollow in the middle" pattern. Upstream foundation models and chips attract capital; downstream use-case demand is robust. But the midstream lacks an engineering platform capable of converting domain expertise into reliable agents.

Another signal worth watching is the explosion of OPCs (One-Person Companies). One-person limited liability companies nationwide have surpassed 16 million, accounting for 27.4% of all enterprises. 2026 has been dubbed "the Year of the OPC," with over 20 cities rolling out dedicated OPC support policies. The core need for these one-person companies is to replace traditional teams with AI Agents—achieving "one person, one army"—but LangChain requires coding skills, and Dify's workflows don't support true autonomous decision-making.

Xiaomi's position is unique: it has the MiMo foundation model, ranked first globally among open-source models on the Agent Index; it has an ecosystem of 1 billion IoT devices and 746 million monthly active users; and it has dramatically competitive API costs after the 99% price cut. But without an Agent-building platform that non-technical users can actually pick up and use, none of these resources can be fully converted into commercial value.

SoloEngine is the key to solving this problem. MiMo provides the Agent's "brain"; SoloEngine provides the Agent's "hands and feet." Together, they elevate Xiaomi from the "building models" strategic phase to the "building platforms" and "building services" phase.

Xiaomi's ecosystem advantages are amplified further through SoloEngine: MiMo's model capabilities, the 99% cheaper API costs, 1 billion IoT devices, the Agent ecosystem platform, the miclaw phone Agent—these resources are woven together by SoloEngine into an ecosystem moat that other platforms can't easily replicate.

While OpenAI is still locking AgentKit into the GPT-5 ecosystem, Xiaomi has already driven the barrier to building Agents down to zero with the MiMo-plus-SoloEngine combination.

At the epicenter of this seismic shift in the Agent landscape stands Xiaomi and its newly launched SoloEngine.

As SoloEngine's own tagline puts it: No Workflow. No orchestration code. Just Agents that get things done.

DEV Community: Dufrence

Lei Jun's Next Move After LLM, Reshapes the Agent Landscape

I. Xiaomi's AI Year in Review: From 7B to a Trillion-Parameter Powerhouse

II. How SoloEngine Is Reshaping the Agent Landscape

III. Why Xiaomi Is Entering the Agent Space