Google has upgraded NotebookLM with a new reasoning engine, expanded file output options, and a more flexible research workflow, giving the AI notebook tool a broader set of capabilities for handling complex projects.

Smarter reasoning, powered by Gemini 3.5

The upgrade moves NotebookLM onto Gemini 3.5 and Antigravity, Google’s coding model, bringing more accurate responses and greater transparency into how the tool reaches its conclusions.

Google says each notebook now runs on a dedicated cloud computer that can write and execute code, backed by over 100 curated software skills. Google’s own benchmarks put the new system at a 65%-plus win rate over the previous version across five core evaluation categories, with particularly strong gains in large document analysis (69.9%) and web research (78.2%).

More output formats, and a lower barrier to start

NotebookLM can now generate a wider range of downloadable files directly from the studio panel, including PDFs, Word documents, Excel spreadsheets, PowerPoint decks, CSVs, data visualizations, and images. Users can provide detailed formatting instructions before generation and request edits afterward.

On the research side, NotebookLM no longer requires a fully formed source library to get started. Users can open a notebook with a rough idea or question, and the tool will use Google Search to surface relevant sources and help build out the repository. Notably, users retain full control over which sources are added, and all sources remain clearly attributed throughout the notebook.

The updates are live now for Google AI Ultra subscribers and Workspace business customers on AI Ultra Access and AI Expanded Access plans, with broader availability planned for later.