NotebookLM gets Gemini 3.5, cloud code execution via Antigravity, and web-first research starting from a chat. Available now for Google AI Ultra and Workspace users.
Google pushed a significant upgrade to NotebookLM on Monday, replacing its model backend with Gemini 3.5 and connecting the research tool to a live cloud computer that can write and execute code during research sessions. The Verge reported the changes, which Google described as "across the board" in a company blog post.
The update lands first for AI Ultra plan subscribers and Workspace customers. A broader rollout was promised but not dated.
NotebookLM launched in 2023 as a document-grounded alternative to general-purpose artificial intelligence chat. Users upload notes, PDFs, and video links, then interrogate that material without the model wandering beyond its stated sources. Monday's update keeps that architecture and adds connectivity to Antigravity, Google's agentic coding platform, which ties each notebook to what the company calls a "secure cloud computer."
The cloud computer
That infrastructure change is the most consequential part of the update. NotebookLM can now write code, run it in that cloud environment, and return results within the same session. Exports have expanded well beyond plain text: users can pull out PDFs, PNG and SVG data visualizations, images generated by Nano Banana (Google's own image generation system), Excel spreadsheets, PowerPoint presentations, and CSV files. The tool is evolving from reading companion toward a lightweight research workstation.
The update also rewrites how a session can begin. The Verge notes that users previously had to upload or link source documents before the AI would engage substantively. Now NotebookLM can query Google Search from the first message, building out a source set as the user types. The change extends an existing "discover" feature that surfaced relevant web resources.
Finding sources first
This sourcing shift is subtle but strategically significant. Tools like Perplexity have long offered web-first sessions, and NotebookLM's requirement to pre-assemble a document library created friction, especially for exploratory research where the relevant sources are not yet known. Letting artificial intelligence handle the initial source triage lowers the entry cost.
Whether it preserves NotebookLM's reliability record is less certain. The app's reputation rests on tight attribution: responses trace to user-verified material, not to whatever a search algorithm surfaces. Handing source selection to Google Search introduces quality variance that the original design explicitly avoided, and The Verge frames the change as an extension of the discover feature rather than a fundamental shift, though the architectural difference is real.
Context
NotebookLM became a breakout product partly because of Audio Overview, its podcast-style document summary feature, and partly because knowledge workers wanted artificial intelligence tools with traceable sources. The Gemini 3.5 swap targets accuracy and reliability within that existing framework, arriving as successive model generations have improved measurably on reasoning benchmarks.
Competitive pressure is significant. Microsoft has embedded document intelligence throughout Copilot and its Office suite. Startups targeting knowledge workers multiply each quarter. Keeping NotebookLM current with a new model and a cloud compute layer is table stakes positioning, not a lasting differentiator. The Verge notes the new capabilities are restricted to paying tiers, with free users waiting on an expansion that Google has not yet scheduled.
The Antigravity integration may have implications beyond this single product. If Google standardizes that agentic coding layer across Workspace, giving AI tools a live compute environment as a default could reshape how document-heavy applications handle complex tasks at scale.
For Google, the real test will be whether it can deliver these capabilities to its broader user base without compromising the source discipline that made NotebookLM worth trusting.
---
Frequently asked questions
What model does NotebookLM now use?
Google upgraded NotebookLM to Gemini 3.5. The company says the newer model provides more accurate and reliable responses than the prior version.
What is Antigravity?
Antigravity is Google's agentic coding platform. NotebookLM now runs on it, giving each notebook a cloud computer that can write and execute code, then deliver results as PDFs, spreadsheets, visualizations, or presentation files.
Who can access the new features right now?
The Gemini 3.5 upgrade and cloud computer capabilities are live for Google AI Ultra plan subscribers and Workspace customers. Google has signaled a broader rollout but has not committed to a date.
Does NotebookLM still require users to upload documents before starting?
No longer mandatory. The update lets users start with a question, and NotebookLM will use Google Search to find relevant sources. Uploading specific documents remains an option for users who want tighter control over what the AI can reference.