Meituan AI Browser is Here: Over 10 Top Models to Do the Work for You, Free with a Bonus Agent Plugin

If AI products have been scrambling for the "chat box" over the past year, Meituan has placed its entry point in a more ancient, yet much more frequent location: the browser.

On June 9, the GN06 team under Meituan announced the official launch of the AI-native browser Tabbit 1.0. Public reports indicate that Tabbit is not simply adding a chat sidebar to a browser, but rather integrating web browsing, global search, AI dialogue, multi-model invocation, and Agent automated task execution into a single workspace.

The interesting part of this is: Meituan didn't start by building a "universal AI App," but chose to enter from the browser.

Because for many white-collar workers, students, and content creators, the browser is where work actually happens. Looking up information, viewing web pages, opening backends, organizing documents, downloading files, comparing information, and filling out forms—almost all of it starts with the browser.

Therefore, what Tabbit truly wants to capture is not just the browser market, but the "work entry point of the AI era."

1. Browsers No Longer Just Open Web Pages; They Start Working for You

Traditional browsers solve the question of "where to look."

Search engines tell you where the links are, and the browser is responsible for opening the web pages. As for how to summarize, copy, organize, fill out forms, or generate documents after reading a web page, users had to do it all themselves in the past.

But AI browsers want to solve the question of "what to do next."

According to public reports, Tabbit is an AI entry point in the form of a browser. After users input their needs, it can automatically execute complex tasks across software and web pages. For example, in an HR scenario, from screening resumes to generating a PPT, one used to have to switch back and forth between recruitment systems, spreadsheets, documents, and presentation software. Now, it can be handed over to an Agent to complete with a single natural language command.

This means the positioning of the browser has changed:

It is no longer just a web page container, but a task executor.

Users are not just "opening a page," but initiating a workflow within the browser.

This is also why AI browsers have become the focus of a new round of entry point competition. Compared to standalone chatbots, browsers naturally connect to web pages, files, accounts, online services, and various business systems. As long as the Agent can stably understand pages, invoke tools, and execute actions, the browser could become the most natural AI workstation.

2. Over 10 Models Are Not a Gimmick; the Core is Returning Model Selection Power to Users

Another key point of Tabbit is its multiple models.

Public information shows that Tabbit 1.0 has built-in multiple leading domestic large models and will connect to new model APIs in real-time. According to the Economic Information Daily, Tabbit has integrated models such as LongCat, DeepSeek, Zhipu GLM, and Kimi; other reports mention that the product supports over 10 leading domestic large models and supports multi-model comparison.

This in itself is not hard to understand.

Because in real work scenarios, no single model is always the best fit for all tasks.

Writing copy might flow better with one model, while reasoning might be more stable with another. Long-text summarization, web reading, code processing, spreadsheet analysis, and information retrieval all have different requirements for model capabilities.

If an AI application is strongly bound to a single model, users can only accept one default answer. The value of a multi-model browser is allowing users to switch models, compare results, and even use different models to cross-verify within the same workstation.

Public reports also mention that Tabbit's public beta data shows that over 60% of active users will switch base models according to different scenarios, or use the "multi-model comparison" feature to gain more inspiration.

This indicates a trend: users don't lack understanding of models; rather, they need a lower-barrier model scheduling interface.

In the past, model selection was the business of developers, platforms, and product managers; now, it is starting to become part of the ordinary user's workflow.

3. Free is Not the End Goal; Agents Are the Key to Commercialization

"Free" is the most easily spotted keyword in this wave of communication.

According to multiple public reports, Tabbit's core functions for general users—such as basic dialogue, web reading, and popular tricks—will be permanently free. At the same time, for high-frequency Agent automation invocation and advanced customization scenarios, Tabbit will explore differentiated subscription models to balance computing power costs.

This actually makes a lot of sense.

The basic capabilities of an AI browser can be free because it needs to quickly build user habits; but the truly expensive part is the Agent executing complex tasks at a high frequency.

A single web reading might just be one Q&A session; but a single Agent task might involve multiple rounds of web page understanding, multiple model invocations, page operations, file generation, error retries, and context memory behind the scenes.

Such tasks consume not just Tokens, but also the stability, permissions, security, and product experience within the execution chain.

Therefore, Tabbit's commercialization logic might not be selling "chat times," but selling "the ability to automatically complete work."

In other words, what users will be willing to pay for in the future is not what the AI answered, but how much repetitive labor it saved them from doing.

4. From the Address Bar to the Dialog Box, and Then to the Agent

The Economic Information Daily report mentioned that Tabbit's head, Liu Jiong, stated that based on user feedback, Tabbit has undergone a leap from the "address bar" to the "search box," then to the "dialog box," and ultimately evolved into an "Agent."

This sentence is worth breaking down.

In the address bar era, the browser was the entry point; in the search box era, the browser connected to information; in the dialog box era, the browser began to understand user intent; in the Agent era, the browser must execute actions on behalf of the user.

This is the biggest difference between AI browsers and traditional browsers.

Traditional browsers organize web pages, while AI browsers attempt to organize tasks.

The newly added memory feature in Tabbit 1.0 is also supplementing this capability. Public reports show that it can record user preferences, backgrounds, and other important information to form "callable memory," automatically adapting to the user's reply style and reducing invalid dialogues and repetitive operations.

If the browser is seen as a workstation, memory is the user's work context; if the Agent is seen as the executor, memory is the prerequisite for it to understand user habits.

5. Why is Meituan Making an AI Browser?

On the surface, it seems a bit of a leap for Meituan to make a browser.

But if you place it within the competition for AI application entry points, it's not hard to understand.

Meituan itself is a company strong in products, fulfillment, and density. In the past, it dispatched merchants, riders, users, and local life services; in the AI era, complex tasks also require dispatching: models, web pages, files, tools, permissions, processes, and results.

The browser happens to be a sufficiently universal entry point.

It is not limited by scenarios like a single vertical App, nor is it too far removed from real web pages and business systems like a pure chatbot. It connects to the work pages users are using every day, making it easier to support Agent task execution.

Therefore, Meituan making an AI browser is essentially not about replicating a traditional browser, but about finding a high-frequency, low-friction AI entry point that can support complex workflows.

6. The Real Highlight: AI Products Are Shifting from "Answering Questions" to "Completing Tasks"

The significance of Tabbit 1.0 does not lie in having one more browser.

It is more like a signal: AI applications are entering the "execution tool" stage from the "Q&A tool" stage.

In the past, AI helped you write a paragraph, summarize an article, or answer a question; now, AI is starting to help you collect information across web pages, organize materials, generate files, fill out forms, compare, review, and deliver results.

This is a wake-up call for all office products:

The entry point of the future may not necessarily be an App icon or a search box, but rather an Agent workstation that can understand tasks, dispatch models, invoke tools, and execute continuously.

The browser is simply one of the most natural shells for this workstation.

Conclusion

The highlights of Meituan's AI browser Tabbit 1.0 can be summarized in three sentences:

First, it pushes the browser from a "web page entry point" to an "AI work entry point";

Second, through multi-model integration, it allows users to choose more suitable models in real scenarios;

Third, it uses Agent capabilities to attempt the automation of complex tasks across web pages and software.

Free is certainly important, but what truly determines whether such products can survive is not how large the free quota is, but whether it can stably complete the most trivial and time-consuming tasks for users.

If the AI browser succeeds, the next battle for the office entry point might no longer take place in the chat box, but in the first browser window everyone opens every day.

Reference Information:

[ref_1] Economic Information Daily: "Meituan Lays Out AI Browser, Tabbit 1.0 Launches"

[ref_2] The Paper: Reports on Meituan releasing the AI browser Tabbit 1.0, executing tasks across software and web pages, and the standard edition being permanently free

[ref_3] Pacific Tech: Reports on Tabbit 1.0 integrating multiple large models, improvements in Agent task success rates, availability on Windows/macOS, and mobile testing

[ref_4] Sina Finance / Ebrun: Reports on Tabbit 1.0, multi-model switching, memory features, an open Skill ecosystem, and the exploration of a subscription model