Xiaomi open-sources MiMo Code, while highlighting SoloEngine as its broader agent platform push

MiMo Code Goes Open Source, but Xiaomi's Real Ambition Is MiMo SoloEngine

In the early hours of June 11, Xiaomi officially released and open-sourced MiMo Code. Released under the MIT license, it comes with a free built-in MiMo-V2.5 model and scored 62% on SWE-Bench Pro—5 percentage points higher than Claude Code. The media went wild. Developers were ecstatic. But if you only noticed MiMo Code, you missed the most critical move in Xiaomi's grand strategy. MiMo Code is built for developers. Xiaomi's real ambition is MiMo SoloEngine—a low-code platform that enables every industry to build Agentic AI systems. 1. How Powerful Is MiMo Code? Let's start with MiMo Code, as it is indeed Xiaomi's heavyweight weapon in the Agent space. Persistent Memory System—A triple mechanism comprising project memory, session checkpoints, and task progress. The main Agent focuses on execution while an independent SubAgent handles all documentation. Context window filling up? The SubAgent automatically compresses a clean briefing, allowing the main Agent to continue seamlessly. Even after hundreds of long-running conversation rounds, no critical information is lost. Compose Mode—Press the Tab key to switch modes. Provide a simple idea, and the system automatically completes the entire workflow: design → planning → coding → testing → review. Industrial-grade code delivery. /dream Command—Triggered automatically every 7 days, an independent Agent reads historical sessions and existing memory files, performing merging, deduplication, path validation, and compression to converge scattered memories into a compact current state. MiMo Code doesn't start from scratch each time—it grows continuously with accumulated project understanding. Voice Input—Powered by MiMo-V2.5-ASR speech recognition, supporting verbal modification instructions and operational commands without ever touching the keyboard. On SWE-Bench Pro, MiMo Code achieved 62% (Claude Code scored 57%). On Terminal Bench 2, it reached 73% (Claude Code scored 68%). Under identical model conditions, the Agent framework alone accounts for this 5-percentage-point difference. The developers' problem is solved. But what about everyone else? 2. The Gap Between 87% and 10% Amazon Web Services revealed a striking statistic at the AIGC2026 Summit: 87% of enterprises claim to have deployed AI at scale, yet only 10% have genuinely derived value from it. The root cause isn't weak models—MiMo-V2.5-Pro already ranks first globally among open-source models on the Agent Index. Nor is it insufficient compute—after Xiaomi's 99% price reduction, API costs have dropped to electricity-bill-level pricing. The real issue is this: the ability to build Agents remains locked in the hands of developers. Currently, there are only two paths to building AI Agents. The first is low-code Workflow platforms like Dify and n8n. Visual canvas, drag-and-drop nodes, rapid assembly. But their core logic relies on "preset paths"—controlling flows with if/else conditions without supporting true autonomous decision-making. When encountering scenarios outside the presets, the workflow stalls. The second path is code-based development frameworks like LangChain and CrewAI. They support true Agentic AI but require Python programming skills. A lawyer won't use LangChain. An accountant can't configure a ReAct Agent. A marketing specialist doesn't write Python. Low-code platforms are easy to adopt but don't support genuine autonomous decision-making. Code frameworks enable autonomous decision-making but are only accessible to developers. MiMo Code fills the gap for developers. MiMo SoloEngine fills the gap for every other industry. 3. SoloEngine: The True Agent Platform for Everyone SoloEngine is the first low-code Agentic AI development platform. Its mission is clear: enable people in every industry to build their own Agentic AI systems, just as developers use MiMo Code for Vibe Coding. How does it achieve this? SoloEngine encapsulates the ReAct architecture, Tool calling, MCP protocol, Skills, and SubAgents entirely in the backend. Users don't need to understand these technical terms—just open a browser, drag Agents onto the canvas, connect their collaboration relationships, configure the required tools, and click Run. The backend automatically compiles everything into a dedicated Agentic AI system. This system is just as powerful as MiMo Code. Each Agent runs a ReAct loop (Think → Act → Observe → Repeat), making real-time judgments based on current conditions. When encountering unexpected situations, it adjusts its strategy autonomously. When discovering better approaches, it proactively switches paths. The entire process requires no if/else statements and no preset paths. Progressive Disclosure—Tools, Skills, and MCP protocols load on demand, reducing token consumption by over 85%. Unified Adaptation Layer—covering all major models including OpenAI, Anthropic, Ollama, MiMo, DeepSeek, Tongyi Qianwen (Alibaba), and Zhipu. One-Click Packaging—assembled Agent teams can be packaged into complete products for self-deployment or distribution. Open-sourced under Apache 2.0, completely free. 4. The OPC Era: SoloEngine's True Battlefield Nationwide, One-Person Companies (OPCs) have surpassed 16 million, accounting for 27.4% of all enterprises. 2026 has been dubbed the "Year of OPC," with over 20 cities introducing dedicated OPC support policies. The core demand of these "one-person companies" is to replace traditional teams with AI Agents, achieving "one-person armies." But LangChain requires programming skills, and Dify's Workflows don't support genuine autonomous decision-making. SoloEngine was built precisely for this era. A cross-border e-commerce operator builds an Agent team with SoloEngine: a Product Analysis Agent (automatically analyzing competitor data), a Copywriting Agent (generating product descriptions), and a Customer Service Agent (automatically handling customer inquiries). Three Agents collaborating—running 6 online stores single-handedly is no longer a myth. A lawyer builds a legal affairs Agent team with SoloEngine: a Case File Analysis Agent (automatically extracting key information from case files), a Legal Research Agent (automatically matching relevant regulations), and a Document Generation Agent (automatically drafting legal document templates). The lawyer only needs to review and sign. An independent developer builds a full-stack development Agent team with SoloEngine: a Requirements Analysis Agent, an Architecture Design Agent, a Code Implementation Agent, and a Testing Agent. One person accomplishing what once required an entire team. SoloEngine transforms Vibe Coding into Vibe Everything. 5. Xiaomi's True Ambition Xiaomi's pace of expansion in AI has exceeded most expectations. In April 2025, MiMo-7B was open-sourced. In March 2026, MiMo-V2-Pro launched, and the miclaw mobile Agent began closed beta the same month. April saw the MiMo-V2.5 series open-sourced and the Agent ecosystem platform enter public beta. In May, MiMo-V2.5-Pro claimed the top spot globally among open-source models, with API prices slashed by 99%. June brought MiMo Code's release and open-sourcing, alongside SoloEngine's simultaneous open-source launch. From a 7-billion-parameter model to a trillion-parameter flagship, from follower to frontrunner—Xiaomi achieved this in less than a year. But no matter how powerful large models are, they are essentially just a "brain." What Xiaomi aims to do next is equip this brain with "hands and feet." MiMo Code provides hands and feet for developers. SoloEngine provides hands and feet for everyone. MiMo provides the "brain" for Agents, while SoloEngine provides the "building platform" for Agents. Together, they elevate Xiaomi's strategy from "building models" to "building platforms" and "building services." Xiaomi's ecosystem advantages are further amplified through SoloEngine: MiMo's model capabilities, 99%-reduced API costs, 1 billion IoT devices, the Agent ecosystem platform, and the miclaw mobile Agent—these resources are interconnected by SoloEngine, forming an ecosystem moat that other platforms cannot easily replicate. MiMo Code is open source, and developers are celebrating. But Xiaomi's true ambition is to give every industry its own "Claude Code." This ambition has a name: MiMo SoloEngine. Open Source · Apache 2.0 · github.com/Sh4r1ock/SoloEngine

15 hours ago

MiMo Code Enters the Coding Agent Space — But the Xiaomi AI Team's Agent Story Started with SoloEngine

On the morning of June 11, Xiaomi's MiMo tech team officially released and open-sourced MiMo Code V0.1.0. The official positioning is clear: Xiaomi's first entry into the Coding Agent space. It's an experimental AI coding assistant that runs in the terminal. Built on top of OpenCode, open-sourced under the MIT license, with a time-limited free MiMo-V2.5 multimodal model. It ships with an original persistent memory system — a triple mechanism of project memory, session checkpoints, and task progress — where the main Agent focuses on the work and logging is entirely handled by an independent SubAgent. Window getting full? The SubAgent automatically compresses a clean briefing, and the main Agent carries on. Hundreds of turns in a long-running session, and no key information is lost. Combined with Compose mode — press Tab to switch, give a simple idea, and the system automatically runs the full pipeline: design → plan → code → test → review. SWE-Bench Pro score: 62%, five points higher than Claude Code. Xiaomi's official statement: MiMo Code starts with programming, but it doesn't stop at programming. It's not just a handy AI coding assistant — it's an AI teammate that lives in your terminal and gets smarter the more you use it. But if you think this is the Xiaomi AI team's first time building an Agent, you couldn't be more wrong. 1. The Project from Eight Days Ago On June 3, Xiaomi open-sourced a project. No launch event, no press release, no Lei Jun retweet. It quietly appeared on GitHub. The project is called SoloEngine. Apache 2.0 license, completely free. Its positioning is far bigger than MiMo Code's — the first low-code Agentic AI development platform. MiMo Code's entry point is the terminal — you need to know how to use the command line and understand programming concepts. SoloEngine's entry point is a visual canvas in the browser — drag Agents in, wire them up, click run. The backend automatically compiles everything into a runnable Agentic AI system. No code needed. No need to understand technical terms like ReAct, MCP, or SubAgent. The Xiaomi AI team's debut in the Coding Agent space was MiMo Code. But their real layout in the Agent space started with SoloEngine. 2. SoloEngine: Giving Every Industry Its Own Agent SoloEngine solves the same problem as MiMo Code — enabling Agentic AI to handle long-running tasks, make autonomous decisions, and support multi-Agent collaboration — but the entry point is completely different. MiMo Code is for developers. SoloEngine is for everyone. A lawyer opens SoloEngine, drags three Agents onto the canvas — "Contract Review Agent," "Legal Statute Search Agent," "Risk Flagging Agent." Wires up the collaboration relationships, configures the tools, clicks run. 30 minutes later, a contract review report is automatically generated with 37 risk points flagged. A cross-border e-commerce operations manager builds an Agent team on SoloEngine — one product analysis Agent, one copywriting Agent, one customer service Agent. Three Agents collaborating, one person running six online stores. An indie developer builds a full-stack development Agent team on SoloEngine — one requirements analysis Agent, one architecture design Agent, one code implementation Agent, one testing Agent. One person doing the work of an entire team. None of these people can write Python. But they can all build their own Agentic AI systems with SoloEngine. That's the difference between SoloEngine and MiMo Code. MiMo Code lets developers code efficiently with Vibe Coding. SoloEngine lets people in other industries work efficiently with Vibe Everything. 3. Same Team, Same Technical Philosophy SoloEngine and MiMo Code come from the same AI team. Though they target different users, their underlying technical philosophy is the same. Real Agentic AI is not a preset Workflow. Each Agent in MiMo Code runs in an autonomous decision-making loop — it decides on its own what to do next, which tools to call, and how to adjust its strategy. Each Agent in SoloEngine also 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. MiMo Code uses a persistent memory system to solve long-session "amnesia." SoloEngine uses multi-Agent collaboration to distribute context pressure — each SubAgent has its own independent context window, loading only the instructions and tools it needs. Progressive disclosure cuts Token consumption by over 85%. MiMo Code uses Compose mode to go from idea to code in one click. SoloEngine uses topological compilation to go from visual design to a runnable Agent team in one click. Different entry points, same philosophy. Different products, same team. 4. Why SoloEngine Came First June 3: SoloEngine. June 11: MiMo Code. First the platform for everyone, then the tool for developers. There's a clear strategic logic behind this sequence. The Xiaomi AI team saw a core contradiction: 87% of enterprises claim to have deployed AI at scale, but only 10% have actually gotten value from it. The gap isn't because models aren't powerful enough — MiMo-V2.5-Pro already ranks first on the global open-source model Agent index, and after a 99% API price cut, costs have dropped to "pennies." The real problem: the ability to build Agents is locked inside the hands of developers. Low-code Workflow platforms (Dify, n8n) are easy to pick up, but don't support true autonomous decision-making. Code-based development frameworks (LangChain, CrewAI) support autonomous decision-making, but only developers can use them. SoloEngine fills exactly this gap — zero code plus true Agentic AI. The Xiaomi AI team released SoloEngine first because the primary problem they're solving isn't "how developers write code" — it's "how every industry uses Agents." MiMo Code is icing on the cake; SoloEngine is the one that actually matters. 5. Xiaomi AI's Agent Map From SoloEngine on June 3 to MiMo Code on June 11, the Xiaomi AI team completed a two-pronged layout in the Agent space in a single week: Product Positioning Release Date Target Users License MiMo SoloEngine Low-code Agentic AI platform June 3 All industry professionals Apache 2.0 MiMo Code Terminal-based AI coding assistant June 11 Developers MIT miclaw Mobile AI Agent March (closed beta) Mobile users - Agent Ecosystem Platform Agent distribution marketplace April (open beta) Developers / users - Both product lines share the same model foundation — MiMo. The same technical philosophy — real Agentic AI. The same goal — letting AI complete tasks autonomously, not follow preset paths. The Xiaomi AI team is telling the market with two products: we're not building a single tool — we're building a complete Agent ecosystem. MiMo Code starts with programming, but it doesn't stop at programming. SoloEngine starts with Agents, but it doesn't stop at Agents. Developers already have MiMo Code. For everyone else — it's time to meet MiMo SoloEngine. Open source · Apache 2.0 · github.com/Sh4r1ock/SoloEngine

15 hours ago