Meituan recently released its AI browser, Tabbit 1.0. Seems like Perplexity’s Comet isn’t much of a barrier — anyone can whip up a similar (or even better) product. Typical Indian hype, short on delivery… https://meituan.com/news/NN260612179003187…

Meituan Enters AI Browser Space with Tabbit 1.0 Launch

Source: https://www.meituan.com/news/NN260612179003187 Meituan’s GN06 team announced today the official launch of its AI-native browser, Tabbit 1.0. Described as a browser-based AI entry point, Tabbit integrates multiple top-tier large language models. When users input a request, Tabbit can automatically execute complex tasks that span software and web pages.

According to Liu Jiong, head of the Tabbit project, the browser entered public beta on March 2 this year. During the beta period, Tabbit received weekly updates based on user feedback, adding over 100 new features. The Agent task success rate improved from 53.1% in March to the current 91.8%. For example, in an HR work scenario — from screening resumes to generating PPTs — tasks that previously required multiple software tools can now be completed with a single command.

“Based on user feedback, Tabbit has evolved from an ‘address bar’ to a ‘search box’, then to a ‘dialog box’, and finally into an ‘intelligent agent’,” Liu said. This evolution is reflected not only in technical metrics but also in a profound shift in user behavior. May data shows that the average monthly token usage per user has reached 8.53 million, indicating that users are consistently and frequently applying Tabbit to heavier task processing and workflows.

To build a personalized “AI entry point,” the newly launched version 1.0 adds a memory function that continuously records user preferences, background, and other important information, forming “callable memories.” It automatically adapts to the user’s reply style, reducing ineffective dialogues and actions.

Addressing the common pain point of “AI applications being strongly tied to a single model,” Tabbit 1.0 includes several leading domestic large models, such as LongCat, DeepSeek, Zhipu GLM, and Kimi, and will connect to new model APIs in real time. Beta data shows that over 60% of active users switch base models according to different scenarios or use the “multi-model comparison” feature to cross-verify and gain more inspiration.

According to Liu, core features for general users — such as basic conversation, web reading, and common tips — will remain permanently free. For high-frequency Agent automation calls and advanced customized scenarios, Tabbit will explore a differentiated subscription model to balance computing costs.

Currently, versions for Windows and macOS are available simultaneously, and the mobile version has also entered testing.