Three Routes for AI Browsers and Tabbit 1.0

机器之心 · wechat · 2026-07-12

Synced outlines three main paths for AI browsers in 2026, using Tabbit 1.0 by Meituan's GN06 team to explore their evolution.

Three Categories of AI Browsers

Traditional Engine + AI Sidebar: Retains the original browser architecture, overlaying AI capabilities via sidebars or extensions (e.g., Edge + Copilot, Brave + Leo).
Research/Agent-Oriented: Built on engines like Chromium, embedding visual multimodal comprehension and cross-tab context memory deeper into the browsing experience (e.g., Perplexity Comet, ChatGPT Atlas).
AI-Native: Integrates Browser + Agent + Workflow directly into the runtime, fundamentally orchestrating task scheduling, multi-model collaboration, and workflows (e.g., Tabbit, Dia, Fellou).

Features of Tabbit 1.0

Interaction entry points have evolved from the address bar to search boxes, chat boxes, and now autonomous agents.
Task scopes have expanded from Q&A and search to cross-web data extraction, form filling, and file generation.
Within 100 days of beta testing, the built-in Agent task success rate improved from 53.1% to 91.8%, with monthly average Token consumption per user reaching 8.53 million.
Tabbit integrates multiple models including LongCat, DeepSeek, GLM, and Kimi, supporting real-time API integration and showcasing a switchable, schedulable multi-model architecture.

The article further discusses how moving AI from the sidebar into the browser's core will transform an agent's execution scope, context awareness, and permission boundaries.

Original post →

Three Routes for AI Browsers and Tabbit 1.0

More from Companies & People