NotebookLM Gemini Google Antigravity AI tools

NotebookLM Gets Gemini 3.5, Antigravity Integration, and Actual File Output

17 June 2026

Google's NotebookLM is receiving a major update, upgrading to the Gemini 3.5 Flash model, which delivered a 65% win rate over the older Gemini 3.1 model across key evaluation areas. The update also introduces Antigravity integration, giving NotebookLM a "cloud computer" with over 100 software skills to run code and build workflows, along with the ability to generate a variety of file types such as PDFs, slides, charts, and images. Additionally, NotebookLM can now automatically find and import web sources directly from the chat interface, with the updates rolling out first to AI Ultra subscribers before reaching other users.

Google's NotebookLM has quietly become one of the more useful things the company has shipped in recent years, which is presumably why it hasn't been quietly euthanised like so many of its stablemates. Today it gets a meaningful upgrade: Gemini 3.5 Flash under the hood, code execution via Antigravity, expanded file format support, and smarter web source discovery.

Gemini 3.5 Flash was announced at Google I/O with the usual promises of faster processing, lower token costs, and equal or better output quality. Whether those claims hold up in practice depends heavily on your use case, but Google has run its own internal benchmarks comparing the old Gemini 3.1 branch against 3.5 across five categories: Accuracy and Quality, Multilingual Support, Large Document Analysis, Document Creation, and Advanced Research. The new model won roughly 65 percent of the head-to-head comparisons. Google is light on methodology details, so take that figure with appropriate salt, but a 65 percent win rate is at least directionally encouraging.

The more interesting addition is Antigravity integration. NotebookLM now has access to what Google is calling a "cloud computer," letting it write and execute code directly in the context of your research. The pitch is that NotebookLM ships with over 100 software skills, theoretically reducing the amount of tab-switching required to build useful workflows around your notebooks. How well that works in practice will depend on whether the code execution is reliable enough to trust, which is a non-trivial question.

NotebookLM has also grown beyond purely text-based outputs. Generated files now land in the Studio Panel alongside existing outputs like audio overviews and quizzes, and you can prompt the tool to revise them after the fact. Supported formats at launch include PNG and SVG charts, PDFs, DOCX, markdown, plain text, CSV, JSON, XLSX, PPTX, and images generated via Nano Banana in PNG, JPG, and GIF formats. More are apparently on the way.

On the sourcing side, NotebookLM can now go and find relevant web sources for you rather than waiting for you to supply them manually. Ask it for more context on a topic and it produces a research report with suggested sources, letting you import all of them or cherry-pick. Those imported pages then become part of the notebook's source pool going forward.

All of this is rolling out now, but not to everyone at once. Gemini 3.5 and the accompanying features are landing first for AI Ultra subscribers and Workspace business customers on AI Ultra Access or AI Expanded Access. Everyone else gets it eventually, timeline unspecified.