As an AI browser, “Task Mode” is a core product feature that Tabbit focuses on polishing. We hope to use a more powerful underlying model to build browser automation capabilities with higher completion rates and more accurate execution effects.
Describe the outcome you want, and Tabbit can move across webpages like a human assistant—extracting information, filling forms, and organizing material across platforms. AI handles the repetitive clicks so you can focus on more creative work.
This workflow puts every part of the underlying model to the test.
It not only needs to "see clearly" the graphics, text, buttons and layout structure in the web page, but also "thinks clearly" about the user's intention. More importantly, it needs to check the current running results in real time during the operation process, so as to dynamically plan and adjust the execution path.
In addition, since a complete browser task often involves multiple rounds of continuous judgments and operations, the response speed, stability and call cost of the model will directly determine the execution efficiency and actual experience of the task. Therefore, when selecting a model, we must not only consider its basic capabilities, but also attach great importance to speed and cost-effectiveness.
Now, Tabbit has released GLM-5.3-Flash. With better task understanding, web page operation and dynamic planning capabilities, as well as a better balance between speed, stability and cost, it will bring a higher completion rate and a smoother and more reliable execution experience to Task Mode.
▎The main model in past scenarios, GLM wins in “cognition” and “planning”
For a long time in the past, Tabbit's core main model has been GLM-5V in browser operation scenarios that rely heavily on multimodal vision. The key reason why it can stand out among many models is that it has taken a key step from "static machine vision" to "dynamic task reasoning".
Most conventional models can only stay at the "clearly visible" image and text recognition stage - they can identify "what's on the page." For Tabbit, the biggest difficulty is understanding what these elements mean and what to do next.
When faced with real web pages with dense information and complex structures, GLM-5V has demonstrated excellent "multimodal comprehensive reasoning capabilities". It can not only accurately analyze the page status, but also closely integrate the user's initial goal, and dynamically adjust the next execution path in real time based on each click feedback during multi-step operations.
It is this autonomous decision-making power that provides solid support for Tabbit’s task flow.
As the Tabbit Task Mode covers more and more scenarios, we have higher expectations for the model's capabilities, speed and cost.
Browser tasks often require the model to complete multiple judgments and operations continuously. If you want users to use it more frequently in daily work, the model not only needs to maintain strong enough visual understanding and reasoning capabilities, but also needs to have faster response speed and higher cost performance.
This coincides with the long-term direction of Z.ai - making cutting-edge intelligence faster, cheaper, and more inclusive, which is also the continued pursuit of the GLM series.
The launch of GLM-5.3-Flash is Z.ai’s practice in this direction. It takes into account capabilities, speed and cost, and also responds to the needs of users in real product scenarios, bringing Tabbit a new option that fits the Task Mode.
▎GLM-5.3-Flash is now online Tabbit
GLM-5.3-Flash is the first native multimodal model of the GLM-5 series, which is newly launched and open sourced by Z.ai. It continues the advantages of the GLM series in visual understanding and task reasoning. It also has faster response speed and higher cost performance, and is more suitable for high-frequency and continuous browser operations in Task Mode.
The model has 320B total parameters and activates 18B per request. Despite being smaller, it outperforms GLM-5.2 (355B-A32B) overall and scores 57 on the Artificial Analysis Intelligence Index, placing it among the world's frontier models.
At the same time, compared with GLM-5.3, the attention calculation amount and KV cache size of this model are reduced by 3.01 times and 4.44 times respectively, and the pricing is only 1/10 of GLM-5.3.
With a new low-cost architecture and 30T token multimodal pre-training corpus, this model can achieve stronger performance with fewer computing resources, making cutting-edge intelligence no longer necessary.
Used in conjunction with the Tabbit Browser, you can more smoothly understand web pages, operate the browser, and advance complex tasks:
Refine key points, check information and generate content after reading long materials;
Collect data across web pages and organize it directly into tables, PPT or reports;
can also more accurately identify the graphic and text content of web pages and complete page operations;
Ability to work collaboratively between web browsing, information retrieval, content understanding, page manipulation and result verification.
Tabbit Browser's conversation and Task Mode are now connected to GLM-5.3-Flash. Click "Switch Model" at the bottom right of the chat box to experience it.
Now you can enjoy a limited-time half-price discount when using it in Tabbit, and the consumption is halved. More conversations and tasks can be completed with the same usage, making cutting-edge intelligence within reach.
Welcome to open Tabbit, select GLM-5.3-Flash in the Task Mode, and experience its performance in visual understanding, task reasoning, and operation execution through real browser tasks.
We would also like to thank the Z.ai team for their support of this experience activity.
For those who haven’t tried Tabbit Browser yet, you can get it through the following methods:
→ PC: Visit Tabbit official website www.tabbit.com or click "Read the original text" below to download and try it for free.
→ Mobile version: Go to major mobile app stores and directly search for "Tabbit" to download and install.
Log in to the same Tabbit account on desktop and mobile, and your bookmarks, tabs, and other browsing data will be seamlessly synchronized.