Google’s NotebookLM, the company’s generative AI research assistant, is receiving what the company describes as its most significant update since launch. The tool, first introduced in 2023 at the dawn of the current AI boom, has survived where many experimental Google products have been shuttered. Today, Google is upgrading NotebookLM to the latest Gemini 3.5 Flash model, expanding its file support, streamlining web source integration, and introducing a new feature called Antigravity — a cloud-based computing environment that allows NotebookLM to write and run code in direct service of user queries.
What is NotebookLM?
NotebookLM (which originally stood for Notebook Language Model) is Google’s attempt to build a personalized AI research assistant that can analyze sources you provide — documents, web pages, PDFs, and more — and answer questions based exclusively on that material. Unlike general-purpose chatbots that rely on their training data, NotebookLM is designed to stay grounded in user-uploaded sources, reducing hallucinations and ensuring relevance. The tool has been popular among researchers, students, and professionals who need to synthesize information from multiple documents.
The Gemini 3.5 Flash upgrade
The centerpiece of this update is the move from the older Gemini 3.1 model to Gemini 3.5 Flash. The new model was first unveiled at Google I/O earlier this year and promised significantly faster inference speeds and improved cost efficiency. Google claims that enterprises worried about token costs — the unit of data processed by large language models — can achieve substantial savings by switching to Flash while maintaining or even improving output quality.
Google conducted internal side-by-side evaluations comparing NotebookLM running on Gemini 3.1 versus Gemini 3.5. The company evaluated the models across what it calls its “top five core evaluation dimensions”: Accuracy and Quality, Multilingual Support, Large Document Analysis, Document Creation, and Advanced Research. In these tests, NotebookLM averaged a 65 percent win rate when using the new model, meaning it outperformed the previous version in nearly two out of every three comparisons. Google did not provide detailed breakdowns of each dimension, but the overall improvement is being attributed to the architectural efficiencies baked into Gemini 3.5 Flash.
For users, the upgrade means faster response times and the ability to handle longer documents without a degradation in performance. Google says NotebookLM can now process source materials that are significantly larger than before, opening up applications in legal analysis, academic research, and business intelligence where large datasets or lengthy contracts must be summarized.
The Antigravity cloud computer
One of the most unusual features in this update is what Google calls “Antigravity.” The name appears to be a playful internal codename that has made its way into the product. Antigravity is described as NotebookLM’s own cloud computer — a built-in environment that can write and execute code based on natural language instructions. For example, a user could ask NotebookLM to “calculate the average growth rate from the sales data in my uploaded spreadsheet” and the system would generate Python or SQL code, run it on the cloud infrastructure, and return the result directly within the chat interface.
Google says NotebookLM will ship with over 100 pre-built software skills that can be chained together to create complex workflows. These skills cover tasks such as data cleaning, statistical analysis, web scraping, and file conversion. By having code execution capabilities embedded directly in the notebook, users no longer need to switch between different applications — they can do everything from within NotebookLM. This moves the tool closer to being a full-fledged data analysis platform rather than just a document summarizer.
The Antigravity feature is reminiscent of tools like OpenAI’s Code Interpreter (now part of ChatGPT), but with a key difference: NotebookLM’s code execution is always grounded in the user’s specific set of sources. That means any analysis or transformation applied to data is done within the context of the uploaded materials, reducing the risk of straying into unverified information.
Expanded file output formats
Another major change is that NotebookLM is no longer limited to text responses. The tool can now generate a variety of file types directly from the chat interface. These outputs are collected in what Google calls the Studio Panel — a dedicated area where infographics, quizzes, audio overviews, charts, and other specialized results appear. Users can also ask NotebookLM to edit or revise these files after they have been created.
The initial set of supported output formats includes:
- Data visualizations and charts in PNG and SVG formats
- Documents in PDF, DOCX, Markdown, and plain text
- Images — Google jokingly calls this “Nano Banana” — in PNG, JPG, and GIF
- Structured data files in CSV and JSON
- Microsoft Excel workbooks (XLSX)
- Microsoft PowerPoint presentations (PPTX)
This expansion transforms NotebookLM from a purely analytical tool into a content creation suite. A researcher could, for instance, upload a set of academic papers, ask NotebookLM to summarize the key findings, and then instruct it to produce a slide deck summarizing those findings — all without leaving the chat interface. Similarly, a business analyst could upload quarterly reports and have NotebookLM generate a set of charts in SVG format suitable for inclusion in a presentation or published report.
Google has indicated that more output types will be added over time, suggesting that the company views NotebookLM as a platform that can handle the entire lifecycle of a research project: from gathering sources, to analysis, to producing final deliverables.
Web source discovery becomes automated
Previously, NotebookLM required users to manually upload or paste URLs for any web-based sources they wanted to include in a notebook. The update changes that by enabling the tool to find relevant web pages on its own. From the chat interface, users can ask NotebookLM to search for sources that match their research topic. The system then returns a “research report” — a list of candidate sources with brief descriptions — and the user can choose to import all or some of them. Once imported, those web pages become part of the notebook’s source set and inform all subsequent interactions. This feature essentially turns NotebookLM into a mini search engine tailored to the specific needs of the user’s project.
By combining automated source discovery with the existing ability to upload documents, NotebookLM now offers a hybrid approach: users can either provide their own materials or let the tool find them. The system still grounds its answers exclusively in the imported sources, so the quality of results depends on the quality of the sources selected. Google recommends that users review the automatically suggested sources before importing them to ensure relevance and accuracy.
Rollout and availability
The update is beginning to roll out today, but not all users will see changes immediately. Google is giving first access to AI Ultra subscribers — the premium tier of its AI service — as well as Workspace business customers with AI Ultra Access and AI Expanded Access. These groups will be the first to experience Gemini 3.5, Antigravity, and the expanded file support. Other Google account holders, including free users of NotebookLM, will receive the updates in the coming weeks.
The staggered rollout is typical for major Google product updates, allowing the company to monitor performance and gather feedback before expanding to the broader user base. Given that NotebookLM relies on computing resources — especially with the new code execution capabilities — Google may also be scaling infrastructure gradually
NotebookLM’s reliance on user-provided or user-approved sources has always been its defining feature, setting it apart from general-purpose AI chatbots that draw from their entire training corpus. With this update, that reliance is slightly relaxed — the tool can now find sources on its own — but the core principle remains: answers are based only on what the notebook contains. This design choice helps mitigate the risk of factual errors and makes NotebookLM particularly attractive for professional use cases where accuracy and source verification are paramount.
As Google continues to integrate its Gemini models into more products, NotebookLM serves as a testbed for features that may eventually find their way into Workspace, Search, or even Android. The addition of code execution and multi-format output suggests that Google is thinking of AI assistants as more than just chat interfaces — they are becoming full-fledged productivity environments. Whether users will embrace the complexity of features like Antigravity remains to be seen, but for now, Google is offering a clear upgrade path for those willing to pay for the latest capabilities.
Source: Ars Technica News