Google is rebuilding NotebookLM on top of Gemini 3.5 and giving every notebook its own secure cloud computer, the company said in a Monday blog post. The upgrade lets the AI note-taking app write and run code on the user's behalf, find its own sources via Google Search, and export research into PDFs, Excel sheets, PowerPoint decks, CSVs, and Nano Banana-generated images. Google described the changes as updates rolling out "across the board."
The shift is more structural than cosmetic. NotebookLM, which launched in 2023 as a way to chat with imported notes and YouTube videos, no longer requires users to bring their own sources to start. A research session can now begin with a question — NotebookLM uses Google Search to surface relevant material, and the user picks which results to import into the notebook.
Under the hood, each notebook is connected to Antigravity, Google's agentic coding platform. That is the part of the announcement that matters most for power users: NotebookLM can now execute code in a sandboxed environment tied to the specific notebook, which is how it generates the new file types and data visualizations.
Key facts
- 01NotebookLM now runs on Gemini 3.5, Google's upgraded model, with what Google calls 'across the board' updates.
- 02Each notebook is wired to a secure cloud computer via Google's Antigravity coding platform, letting NotebookLM write and run code.
- 03Output formats now include PDF, PNG, SVG, JPG, GIF, Excel, PowerPoint, and CSV, plus Nano Banana-generated images.
- 04The upgrade rolls out first to Google AI Ultra and Workspace customers, with broader plans to follow.
- 05NotebookLM launched in 2023 and now lets users start research from a chat prompt instead of importing sources first.
Output formats are the visible payoff. NotebookLM can now produce PDF documents, data visualizations as PNG or SVG, images generated by Google's Nano Banana model in PNG, JPG, or GIF, Excel workbooks, PowerPoint files, and CSV sheets. The app is moving from a chat interface for research into something closer to a research-and-document workspace.
Google said the Gemini 3.5 upgrade allows NotebookLM to respond with "more accurate and reliable information." The discover feature, which helps users find useful resources on the web, is being extended so that source discovery is built into the chat flow rather than sitting as a separate tool.
Distribution is staged. The new version is shipping first to subscribers on Google's AI Ultra plan and to Workspace customers, with the company saying it plans to expand availability to more tiers later. That puts the upgrade behind the most expensive Gemini consumer subscription and the enterprise contract, which is consistent with how Google has been gating its most compute-heavy features.
The cloud-computer-per-notebook design is the same pattern that has been spreading across the agentic tooling market — sandboxed runtimes that let a model execute code, browse, and produce artifacts without the user wiring anything up. Antigravity is Google's house version of that primitive, and bolting it directly into NotebookLM is a faster route to a working agent than asking users to assemble one.
There are real questions Google has not answered. The blog post does not specify per-notebook compute or storage limits, how long the cloud computer persists between sessions, what happens to generated artifacts when a notebook is deleted, or how the system handles sensitive Workspace content under enterprise data-residency rules. Pricing beyond the AI Ultra and Workspace gate is also unspecified, and the announcement did not include benchmark numbers for Gemini 3.5 inside NotebookLM versus the prior model.
Still, the move tightens Google's position in a category it largely defined. NotebookLM has been one of the company's better-received AI products since launch, and turning it into a Gemini 3.5 research agent with code execution and document output narrows the gap between "chat with your notes" and "do the actual work." For the enterprise buyers Google is courting through Workspace, an AI research tool that produces native Excel and PowerPoint files is a more concrete sell than a model leaderboard. The harder test will be whether the cloud computer holds up under sustained, multi-step work — and whether the rest of Google's AI product line gets the same Antigravity treatment as quickly.
Working on something we should cover, or seeing a story we missed? Send leads, documents, or feedback to hello@aichatdaily.com. For sensitive tips, see our secure tips page for Signal and PGP options.
Spotted an error? Email hello@aichatdaily.com with the URL and the issue, or read our full corrections policy.
