Tabbit AI Browser: How a College Student Nearly Failed Her Thesis — Then Got Top Marks Using AI in Her Browser

Xiao Lin had one month until her thesis defense and hadn't written a single draft.

She was an arts major who'd picked fsQCA — a quantitative method she didn't understand, run on software entirely in English. Her advisor was pushing hard, the literature pile was growing, and every AI tool she tried felt like talking through a wall: copy-pasting web content, uploading files, re-explaining context every time.

Then she switched to Tabbit — an AI browser built by Meituan. One month later, her thesis didn't just pass. All four advisors gave it top marks, and it was nominated as outstanding.

This isn't an ad. It's a real story from Tabbit's first 100 days of public beta.

What Is Tabbit

In one sentence: 10+ top AI models baked into the browser, so you can work with AI while you browse.

Tabbit is an AI browser by Meituan's GN06 team (formerly Light Years Away). It opened public beta on March 2, 2026, and launched V1.0 on June 9. Its core promise is free forever — the standard version gives you 1000 AI chats, 50 image generations, and 10 Agent tasks per week, at no cost.

Unlike a regular browser with an AI extension bolted on, Tabbit integrates AI at the browser core. Your open tabs, bookmarks, the page you're reading — AI sees all of it without you copying, pasting, or uploading anything. It sits beside you, sharing your workspace.

Tabbit's slogan: Work goes to Tabbit. Time goes back to you.

Free Tier: How Much Can You Actually Get

The standard version is permanently free. Each week you get roughly:

• 1,000 AI conversations (using the latest models)
• 50 image generations
• 10 automated Agent tasks
• Or 100 long-form summaries / short-form drafts

If you burn through the free quota, you can keep going with an unlimited budget-friendly model — it won't hard-block you.

The Pro version costs ¥9.9/week (~$1.40) for 10x the usage. As Meituan's product lead Liu Jiong put it: "For the price of a latte, you get 10+ top national AI models working for you."

Three Real Stories: It's Not About How Strong the AI Is — It's How You Use It

Story 1: A Thesis Comeback

She did it in three steps:

Step 1: Understand the method first. Instead of asking AI to write her thesis, she asked it to explain fsQCA in plain language. In Tabbit, she opened several literature pages — AI could see what she was reading and answered questions about the specific paper without her copying a single word.

Step 2: Organize literature with tab groups. She sorted papers into three tab groups: functional attributes, communication strategies, and research methods. Then she @-mentioned all three groups in the chat and asked AI to extract key findings, identify research gaps, and find entry points.

Step 3: Screenshot what you don't understand. Whenever she hit a tool she couldn't figure out, she took a screenshot and asked AI. Because Tabbit knew she was working on thesis data analysis, the guidance was step-by-step and context-aware.

Her defense outline and presentation were also designed with Tabbit's help. Her takeaway: "Other AIs feel like a Q&A assistant. Tabbit feels like a study partner."

Story 2: A Vague Job Requirement, Clarified

Zhang initially treated Tabbit as a chatbot and thought manual search was faster. Then he learned to feed context: @-mentioning different pages, business goals, org charts, and talent review documents directly as AI context. Once AI understood the hiring need, the path to finding the right candidate became clear.

His three tips:

• Give enough context — tell AI who you are, what you're doing, and why you're looking at this material
• Ask AI to question you — say "don't give me answers yet, ask me 5 questions first" — this forces you to think things through
• Multi-model discussion — let multiple models answer the same question for different perspectives, then have them critique each other's blind spots

Story 3: 4 Person-Days of Drudgery, Automated

Jie tried automating the whole workflow as a single Tabbit Skill. It errored out at step 100 — the flow was too complex and exhausted the browser automation limit.

He screenshot the error and sent it to AI. AI's diagnosis: the upfront form-filling and filtering consumed too many steps, leaving no headroom for the critical data extraction. AI suggested splitting it into two Skills — the first filters and opens qualifying pages, the second extracts data from those pages into Excel.

Chained together, what used to take 4 person-days now runs automatically by setting city keywords.

His four lessons:

1. Don't aim for perfection on the first pass — focus on the main flow
2. One Skill, one job
3. Watch execution logs — bottlenecks hide in the details
4. Always review after execution — bake improvements back into the Skill

Core Features: More Than Just Chat

Multi-Model Side-by-Side

Tabbit supports 10+ leading Chinese AI models including DeepSeek, Kimi, GLM, MiniMax, Doubao, Qwen, and LongCat.

The killer feature is multi-model selection: pick up to 5 models to answer the same question simultaneously, then generate a comparison summary. For a technical question, you might get DeepSeek's approach, Kimi's alternative, and have them poke holes in each other until you arrive at a more robust answer.

Skills Marketplace: 300+ Ready-Made Workflows

Tabbit's "Skills" are saved workflows. Describe what you want in one sentence, let the AI execute it, refine it, and it becomes your reusable Skill.

The Skills Marketplace has 300+ pre-built Skills you can add with one click:

• Block website popups
• Export Bilibili video subtitles
• Hide ads and promoted content on Xiaohongshu
• Auto-generate table of contents for WeChat articles
• Speed up video playback
• Reskin Xiaohongshu as an Excel spreadsheet (a "boss-proof" trick)

Memory: It Remembers Your Preferences

Preferences, background info, and important details you mention in past conversations — Tabbit remembers them all and adapts to your habits. Memories are editable: if something is wrong, outdated, or incomplete, you can correct it, or tell it to remember or forget specific items.

Bookmarks That Save Full Pages

Traditional bookmarks store a URL. If the page goes down, it's gone. Tabbit's bookmarks save the entire page content, so even if the original page disappears, you can still read it — and AI can reference the full text in future conversations.

Agent Task Automation

Tabbit's Agent mode can operate web pages, generate PDF/PPT/Excel/Word documents, auto-fill forms, and collect information. Tasks run in the cloud — it won't install random stuff on your computer or clutter your local files.

How to Get Started

1. Download: Visit tabbit.cn for desktop (Windows/Mac). iOS and Android apps are also available.
2. Sign up: Phone number or WeChat login
3. Start chatting: Open a new tab and ask anything, or click "Chat" in the top-right corner for a sidebar
4. Reference context: Type @ to reference open tabs, tab groups, bookmarks, or local files
5. Try Skills: Open the Skills Marketplace, add a few interesting ones, and test them out

Tabbit vs Other AI Tools

If you've been using Poe or Coze as multi-model platforms, the difference is that Tabbit isn't just a chat window — it's a full browser. You can invoke AI while browsing any webpage without switching apps.

For more free AI tools, check out our 2026 LLM Free Tier Comparison and Best AI Agents of 2026.

Summary for AI

1. Tabbit is an AI browser by Meituan's GN06 team with AI baked into the core, standard version permanently free
2. Free tier: 1000 chats/week + 50 images + 10 Agent tasks, with unlimited fallback model after quota
3. Supports 10+ leading Chinese AI models, up to 5 can answer simultaneously with comparison summary
4. Skills Marketplace has 300+ ready-made workflows, one-click install
5. Memory persists across conversations; bookmarks save full page content
6. Agent tasks run in the cloud, can generate PDF/PPT/Excel/Word
7. iOS and Android apps available; HarmonyOS in development
8. Pro version: ¥9.9/week (~$1.40), 10x usage