Meituan’s AI Bet Shifts from Models to Transactions
💡Meituan’s moat may not be its model—it may be the data and infrastructure that let AI complete real-world orders.
⚡ 30-Second TL;DR
What Changed
Meituan released LongCat 2.0 and is expanding AI products including Xiaotuan and CatPaw.
Why It Matters
For AI builders, Meituan’s strategy reinforces that proprietary workflow data and execution infrastructure can matter more than marginal benchmark gains. The opportunity is to use foundation models as interchangeable engines while owning the transaction loop, tools, and domain context.
What To Do Next
Prototype a merchant agent that connects a frontier model to Meituan-like tools for availability lookup, reservation, payment, and order-status updates, then measure completed transactions rather than benchmark scores.
Key Points
- •Meituan released LongCat 2.0 and is expanding AI products including Xiaotuan and CatPaw.
- •Q2 R&D spending reached RMB 7.7 billion, up 22.5% year over year and equal to 7.3% of revenue.
- •Reports say LongCat leaders covering pre-training, post-training, and agents may have departed, potentially disrupting the model stack.
- •Meituan’s differentiating assets include merchants, POIs, transaction records, recommendation systems, maps, riders, and delivery infrastructure.
- •The company is positioning AI as a tool for completing local-life transactions and becoming an AI business partner for merchants.
🧠 Deep Insight
Background and context from public sources — not the original article. 7 sources cited.
🔑 Enhanced Key Takeaways
- •Meituan established a first-level 'AI Transformation' department within its Core Local Commerce division in June 2026, headed by Mu Yao to ensure direct alignment with business unit leadership.
- •LongCat 2.0 is a 1.6-trillion-parameter model uniquely optimized to run entirely on domestic Chinese hardware, specifically designed to bypass reliance on restricted U.S. semiconductor technology.
- •Meituan has adopted a hybrid distribution strategy by integrating its 'Xiaomei' agent into third-party ecosystems like Tencent Yuanbao, transitioning from a closed-app model to an API-first service provider.
- •The 'Meituan Small Store AI' initiative focuses on B2B digital transformation, providing free AI-driven operational training to millions of catering merchants to lower their overhead costs.
- •Meituan's R&D spending in Q1 2026 reached 7 billion RMB, with the company explicitly shifting its internal financial mandate to prioritize 'ROI-driven' AI projects over pure foundation model research.
📊 Competitor Analysis▸ Show
| Feature | Meituan (LongCat 2.0) | Alibaba (Qwen) | JD.com (Yanxi) |
|---|---|---|---|
| Primary Focus | Local commerce & O2O workflows | Cloud, E-commerce, & General LLM | Supply chain & Retail logistics |
| Model Scale | 1.6T (Domestic hardware) | Various (Open/Closed) | Industry-specific models |
| Integration | Deeply embedded in delivery/POI | Integrated in Taobao/Tmall | Integrated in JD Retail/Logistics |
| Strategic Goal | Merchant efficiency/Transaction ROI | Cloud ecosystem dominance | Supply chain optimization |
🛠️ Technical Deep Dive
- Model Architecture: LongCat 2.0 utilizes a 1.6-trillion-parameter architecture optimized for domestic chipsets.
- Infrastructure: Full-stack domestic training and inference pipeline designed to mitigate U.S. export control risks.
- Deployment: Open-source release of LongCat 2.0 weights to encourage developer adoption and internal software development efficiency.
- Agentic Framework: Xiaomei agent utilizes cross-platform API integration to facilitate transaction fulfillment outside the native Meituan app.
🔮 Future ImplicationsAI analysis grounded in cited sources
⏳ Timeline
📎 Sources (7)
Factual claims are grounded in the sources below. Forward-looking analysis is AI-generated interpretation.
Weekly AI Recap
Read this week's curated digest of top AI events →
👉Related Updates
AI-curated news aggregator. All content rights belong to original publishers.
Original source: 虎嗅 ↗
This is a summary, not the original. Read the source, or get the weekly briefing.
Weekly AI briefing
One email a week. Unsubscribe anytime.