Meituan Makes a New AI with Chinese Parts

Jun 30, 2026, 13:42
ChinaTechnology
A2

Meituan Makes a New AI with Chinese Parts

美團使用中國零件打造全新 AI

Introduction

Meituan made a new AI model called LongCat-2.0. They used only Chinese computer parts to build it.

美團開發了一款名為 LongCat-2.0 的新 AI 模型，且完全使用中國電腦零件打造。

Main Body

The AI is very big. Meituan used 50,000 Chinese computer chips to teach the AI. They used a special system from Huawei to make the chips work together.

這個 AI 規模非常龐大。美團使用了 5 萬顆中國電腦晶片來訓練該 AI。他們使用了華為的專門系統，使晶片能夠協同工作。

This AI is as good as Google's Gemini AI. But it is not as smart as GPT-5.5. The Chinese software is not as good as NVIDIA software yet.

這個 AI 的能力與 Google 的 Gemini AI 相當。但它不如 GPT-5.5 那麼聰明。中國的軟體目前仍不如 NVIDIA 的軟體。

The USA does not sell some fast chips to China. Now, Meituan can make a big AI without those chips. Meituan also gives money to Chinese chip companies.

美國並不向中國銷售某些高速晶片。現在，美團在沒有這些晶片的情況下也能打造大型 AI。美團同時也向中國晶片公司提供資金。

Conclusion

LongCat-2.0 shows that China can build big AI with its own hardware.

LongCat-2.0 證明了中國可以使用自身的硬體來打造大型 AI。

Vocabulary Learning

⚖️ Comparing Things

In this text, we see a way to say two things are similar or different using as... as. This is perfect for A2 students to describe objects or companies.

How it works:

  • Same level →\rightarrow as [adjective] as
  • Different level →\rightarrow not as [adjective] as

Examples from the text:

  • "This AI is as good as Google's Gemini AI." →\rightarrow (They are equal)
  • "It is not as smart as GPT-5.5." →\rightarrow (GPT-5.5 is smarter)
  • "The Chinese software is not as good as NVIDIA software." →\rightarrow (NVIDIA is better)

Quick Tip: Use this pattern when you want to compare two things without using words like "better" or "worse."

Simple Pattern: A →\rightarrow as →\rightarrow word →\rightarrow as →\rightarrow B

Vocabulary Learning

model (n.)
A version or a type of a computer program
Example:This new AI model can translate many languages.
parts (n.)
Pieces of a larger machine
Example:I need to buy new parts for my computer.
chips (n.)
Small pieces of silicon used in computers to process information
Example:Computer chips make your phone work fast.
system (n.)
A set of things that work together as a whole
Example:The school has a new system for taking attendance.
software (n.)
Programs and other operating information used by a computer
Example:You need to update the software on your laptop.
hardware (n.)
The physical parts of a computer
Example:The keyboard and mouse are computer hardware.
B2

Meituan Launches LongCat-2.0 Using Chinese Computing Hardware

美團推出 LongCat-2.0，全面採用中國計算硬體

Introduction

Meituan has introduced LongCat-2.0, a large language model that was developed using only Chinese hardware for both its training and its operation.

美團推出了 LongCat-2.0，這是一個大語言模型，在其訓練與運行過程中完全使用中國硬體開發。

Main Body

LongCat-2.0 marks a major change in how AI is developed in China. The model features 1.6 trillion parameters and can process a huge amount of data at once. Unlike earlier versions of Chinese AI, which often used foreign hardware for training, LongCat-2.0 was trained on a massive network of 50,000 domestic accelerators. To make this possible, Meituan used specialized super-computers and Huawei's communication software to help the chips work together efficiently.

LongCat-2.0 標誌著中國 AI 開發方式的一個重大轉變。該模型擁有 1.6 兆個參數，能一次處理海量數據。與早期經常使用國外硬體進行訓練的中國 AI 不同，LongCat-2.0 是在一個由 5 萬個國產加速器組成的龐大網絡上訓練而成。為了實現這一目標，美團使用了專用的超級電腦和華為的通信軟體，以協助晶片高效協作。

In terms of performance, the model is as capable as Google's Gemini 3.1 Pro and actually performs better in some specific technical tests. However, it still struggles with complex reasoning when compared to top systems like GPT-5.5. Meituan emphasized that while the hardware works, the domestic software is not yet as advanced as NVIDIA's tools. Furthermore, the company noted that limited memory on each device was a primary problem during the training process.

在性能方面，該模型的能力與 Google 的 Gemini 3.1 Pro 相當，甚至在某些特定的技術測試中表現更佳。然而，與 GPT-5.5 等頂尖系統相比，它在處理複雜推理時仍顯吃力。美團強調，雖然硬體可行，但國產軟體尚未達到 NVIDIA 工具的先進程度。此外，公司指出，每個設備的記憶體限制是訓練過程中的主要問題。

This achievement is strategically important because of US export restrictions on high-end chips. By successfully building a massive model on local hardware, Meituan has shown that China is becoming less dependent on restricted technology. Consequently, Meituan is investing more in local semiconductor companies like MetaX and Moore Threads to ensure they have a reliable, independent supply of computing power.

由於美國對高端晶片實施出口限制，這項成就具有重要的戰略意義。美團成功在本地硬體上構建大規模模型，證明了中國對受限技術的依賴程度正在降低。因此，美團正加大對 MetaX 和摩爾线程 (Moore Threads) 等本地半導體公司的投資，以確保擁有可靠且獨立的算力供應。

Conclusion

LongCat-2.0 proves that it is possible to train world-class AI on Chinese hardware, even though software and memory issues still need to be solved.

LongCat-2.0 證明了即便軟體與記憶體問題仍待解決，使用中國硬體訓練出世界級 AI 仍是可行的。

Vocabulary Learning

The 'Contrast Shift': Moving from A2 to B2

At the A2 level, you probably use 'but' for everything. To reach B2, you need to express how things are different using more sophisticated logical connectors.

Look at these two sentences from the text:

  1. "The model is as capable as Gemini... However, it still struggles with complex reasoning."
  2. "...even though software and memory issues still need to be solved."

The B2 Upgrade Path:

Instead of saying: "The hardware is good but the software is bad" (A2), try these structures:

  • The 'However' Pivot: Use this to start a new sentence. It creates a formal pause that tells the reader a contradiction is coming.

    • Example: "The chips are fast. However, the memory is limited."

  • The 'Even Though' Bridge: Use this to connect two opposite ideas in one sentence. It shows that one fact does not stop the other from being true.

    • Example: "Even though they lack NVIDIA tools, they built a great model."

Quick Vocabulary Bridge

To sound more like a B2 speaker, replace basic verbs with 'Precision Verbs' found in the article:

A2 WordB2 Precision WordWhy it's better
ShowEmphasizeIt means to show something with strength
HelpEnsureIt means to make sure something will happen
UseProcessSpecifically used for data and information

Vocabulary Learning

domestic (adj.)
Existing or occurring inside a particular country; not foreign.
Example:The company decided to use domestic suppliers to reduce shipping costs.
efficiently (adv.)
In a way that achieves maximum productivity with minimum wasted effort or expense.
Example:The new software allows the team to manage their tasks more efficiently.
emphasized (v.)
Gave special importance or prominence to something in speaking or writing.
Example:The manager emphasized the need for accuracy in the final report.
strategically (adv.)
In a way that relates to the achievement of long-term goals or a specific plan.
Example:The company strategically placed its new office near the city center to attract more clients.
restrictions (n.)
Official or official-like rules that limit what you can do or how you can do something.
Example:Due to new government restrictions, travel to the region has become difficult.
dependent (adj.)
Needing someone or something else for support, help, or survival.
Example:The small village is heavily dependent on the weekly market for its food supply.
consequently (adv.)
As a result of something that has happened.
Example:He failed to study for the exam; consequently, he did not pass.
reliable (adj.)
Consistently good in quality or performance; able to be trusted.
Example:We need a reliable internet connection to host the international webinar.
C2

Meituan Implementation of LongCat-2.0 Utilizing Domestic Compute Infrastructure

美團利用國產計算基礎設施實現 LongCat-2.0 部署

Introduction

Meituan has released LongCat-2.0, a large language model developed exclusively using Chinese hardware for both training and inference.

美團發佈了 LongCat-2.0，這是一個在訓練與推理階段均完全使用中國硬體開發的大語言模型。

Main Body

The deployment of LongCat-2.0 represents a significant shift in the technical trajectory of Chinese artificial intelligence. Characterized by 1.6 trillion parameters and a context window of one million tokens, the model utilizes a Mixture-of-Experts (MoE) architecture with approximately 48 billion active parameters. Unlike previous domestic iterations, such as DeepSeek V4-pro, which utilized local silicon solely for inference, LongCat-2.0 completed the computationally intensive pre-training phase on a cluster of 50,000 domestic accelerators. This was achieved through the utilization of large-scale ASIC superpods and Huawei's Collective Communication Library (HCCL) to facilitate inter-chip coordination.

LongCat-2.0 的部署代表了中國人工智慧技術軌跡的一個重大轉向。該模型具有 1.6 兆個參數以及一百萬個 token 的上下文視窗，採用專家混合 (MoE) 架構，約有 480 億個激活參數。與先前僅將國產晶片用於推理的國內版本（如 DeepSeek V4-pro）不同，LongCat-2.0 在一個由 5 萬個國產加速器組成的集群上，完成了運算密集型的預訓練階段。這是透過使用大規模 ASIC superpods 與華為的 Collective Communication Library (HCCL) 來促進晶片間的協調而實現的。

From a performance perspective, the model demonstrates parity with Google's Gemini 3.1 Pro, specifically outperforming it on Terminal-Bench 2.1 and SWE-Bench Pro. However, a performance gap persists when compared to frontier systems such as OpenAI's GPT-5.5 and Anthropic's Opus 4.8 in complex reasoning tasks. The development process revealed systemic vulnerabilities; Meituan noted that the domestic software ecosystem remains underdeveloped relative to NVIDIA's framework, and memory constraints per device were identified as a primary bottleneck during the training phase.

從性能角度來看，該模型表現與 Google 的 Gemini 3.1 Pro 持平，特別是在 Terminal-Bench 2.1 和 SWE-Bench Pro 上超越了後者。然而，在複雜推理任務中，與 OpenAI 的 GPT-5.5 和 Anthropic 的 Opus 4.8 等前沿系統相比，仍存在性能差距。開發過程揭露了系統性脆弱之處；美團指出，國產軟體生態相對於 NVIDIA 的框架仍不夠成熟，且每台設備的記憶體限制被確定為訓練階段的主要瓶頸。

Strategically, this development occurs within the context of United States export restrictions on high-end semiconductors. The successful scaling of a trillion-parameter model on local silicon suggests a diminishing reliance on prohibited hardware. This technical milestone is further supported by Meituan's strategic investments in domestic semiconductor firms, including MetaX and Moore Threads, indicating a broader institutional commitment to compute self-sufficiency.

在戰略上，此次開發發生在美國限制高端半導體出口的背景下。在國產晶片上成功擴展兆級參數模型，表明對禁運硬體的依賴程度正在降低。這一技術里程碑進一步得到了美團對國產半導體公司（包括 MetaX 和摩爾線程）戰略投資的支持，顯示出機構對實現計算自給自足的更廣泛承諾。

Conclusion

LongCat-2.0 establishes the technical viability of frontier-scale AI training on domestic Chinese hardware, despite remaining software and memory limitations.

儘管仍存在軟體與記憶體限制，但 LongCat-2.0 證明了在國產硬體上進行前沿規模 AI 訓練的技術可行性。

Vocabulary Learning

The Architecture of 'Nominalization' as a C2 Power-Tool

To bridge the gap from B2 to C2, a student must move away from action-oriented prose (subject →\rightarrow verb →\rightarrow object) and embrace concept-oriented prose. The provided text is a masterclass in Nominalization: the process of turning verbs or adjectives into nouns to create a dense, objective, and academic register.

🔍 The Linguistic Pivot

Look at the strategic transformation of ideas in the text. A B2 learner describes a process; a C2 writer describes a phenomenon.

  • B2 Approach (Clausal): Meituan developed LongCat-2.0 and this shifted how Chinese AI develops technically.
  • C2 Approach (Nominalized): "The deployment of LongCat-2.0 represents a significant shift in the technical trajectory of Chinese artificial intelligence."

By converting the action ("shifted") into a noun ("shift") and the process ("how it develops") into a complex noun phrase ("technical trajectory"), the writer achieves two things:

  1. Increased Information Density: More conceptual weight is packed into a single sentence.
  2. Objective Distance: The focus shifts from the actor (Meituan) to the implication (the shift in trajectory).

🛠️ Advanced Synthesis: The 'Noun + Prepositional Chain'

C2 mastery is signaled by the ability to chain nouns and modifiers to create highly specific meanings without relying on repetitive verbs.

*"...a diminishing reliance on prohibited hardware."

Instead of saying "they rely less on hardware that is prohibited," the author uses a nominal head (reliance) modified by a participial adjective (diminishing) and a prepositional qualifier (on prohibited hardware).

⚡ Scholarly Application: From Process to Concept

To implement this, replace your primary verbs with their noun counterparts and support them with a 'state-of-being' verb (e.g., represents, indicates, constitutes, signifies).

B2 Verb-CentricC2 Nominalized Transformation
Because the software is underdeveloped...Due to the underdevelopment of the software ecosystem...
They committed to being self-sufficient......indicating a broader institutional commitment to compute self-sufficiency.
It was hard to train because of memory......memory constraints per device were identified as a primary bottleneck...

The C2 Takeaway: Stop describing what is happening and start describing what the occurrence represents.

Vocabulary Learning

trajectory (n.)
The path followed by a projectile or an object moving under the action of given forces; in a figurative sense, the curve of development or progress.
Example:The company's growth trajectory suggests it will dominate the market within five years.
iterations (n.)
The repetition of a process or utterance; in computing, a new version of a piece of hardware or software.
Example:After several iterations of the prototype, the engineers finally eliminated the overheating issue.
parity (n.)
The state or condition of being equal, especially regarding status or pay.
Example:The new domestic model has finally achieved parity with international standards in terms of processing speed.
frontier (adj.)
Referring to the most advanced or cutting-edge stage of development in a particular field.
Example:The research team is pushing the frontier of quantum computing to solve previously impossible equations.
vulnerabilities (n.)
Weaknesses in a system or organization that can be exploited or cause failure.
Example:The security audit revealed several critical vulnerabilities in the network's firewall.
bottleneck (n.)
A situation that causes delay in a process or system, typically because one part is slower than the rest.
Example:Insufficient RAM became the primary bottleneck, preventing the model from training at full speed.
viability (n.)
The ability to work successfully; the capacity of a project or idea to be sustainable and practical.
Example:The pilot program demonstrated the commercial viability of using hydrogen-powered aircraft.
Practice All words in a crossword→