A New Era for NotebookLM

Google has delivered one of its most substantial updates yet to NotebookLM, the generative AI tool that marked one of the company's earliest public forays into consumer AI. Unlike many experimental Google products that have been quietly retired, NotebookLM has not only survived but is evolving rapidly. The latest upgrade brings the Gemini 3.5 Flash model, code execution capabilities via a feature called Antigravity, broader file type support, and smarter web source integration. These changes are set to transform how users interact with their personal research notebooks.

The Power of Gemini 3.5 Flash

At the heart of the update is the transition from the older Gemini 3.1 branch to the newly optimized Gemini 3.5 Flash model. First introduced at Google I/O earlier this year, Gemini 3.5 Flash is designed for speed and efficiency. Google has emphasized that organizations concerned about token costs can achieve significant savings by switching to this model while maintaining—or even improving—output quality. NotebookLM was initially launched in 2023, riding the early wave of the AI boom, and allowed users to analyze specific documents and web pages using Google's latest AI. With the new model, the tool processes queries faster and delivers more accurate results, especially when handling large documents or multilingual content.

Google conducted side-by-side evaluations comparing NotebookLM on Gemini 3.1 versus Gemini 3.5 across five core dimensions: Accuracy and Quality, Multilingual Support, Large Document Analysis, Document Creation, and Advanced Research. The results showed that the updated model achieved an average win rate of 65 percent over its predecessor. While Google has not disclosed the exact methodology or test datasets, the improvement signals meaningful progress in the underlying technology. For users, this means more reliable answers, better handling of non-English sources, and improved performance when summarizing lengthy reports or academic papers.

Antigravity: Code Execution Within Notebooks

One of the most intriguing additions is Antigravity, a feature that allows NotebookLM to write and execute code directly within the research environment. The name is likely a playful internal codename, but the functionality is serious. NotebookLM now comes equipped with over 100 software skills that enable users to build complex workflows without leaving the platform. Previously, such tasks would have required jumping between multiple applications—using a separate code editor, data visualization tool, or spreadsheet program. Now, users can prompt NotebookLM to generate Python scripts, run analyses, produce charts, and even perform data cleaning tasks. The results are integrated seamlessly into the notebook, making it easier to iterate on research questions.

This capability positions NotebookLM closer to tools like Jupyter Notebooks or Google Colab, but with a more conversational interface. Instead of writing code line by line, users can describe what they want—such as “create a scatter plot of the data from my uploaded CSV file”—and NotebookLM will generate and execute the appropriate script. The system handles errors automatically and can adjust based on follow-up instructions. For researchers and analysts who are not professional programmers, this lowers the barrier to performing sophisticated data operations.

Expanded Output Formats

Another significant enhancement is the expansion beyond plain text outputs. NotebookLM can now generate a variety of file types directly from the Studio Panel, including infographics, quizzes, audio overviews, and other specialized formats. Users can request these outputs using natural language and later edit them. Google has announced initial support for the following file types:

• Data visualizations and charts (PNG, SVG)
• Documents (PDF, DOCX, Markdown, TXT)
• Images with Nano Banana (PNG, JPG, GIF)
• Structured data (CSV, JSON)
• Microsoft Excel (XLSX)
• Microsoft PowerPoint (PPTX)

The inclusion of PowerPoint and Excel formats makes NotebookML useful for business presentations and data sharing. The “Nano Banana” image feature appears to be an in-house image generation model, possibly a compact version of Imagen. The ability to generate quizzes and audio overviews (via the existing Audio Overviews feature) further diversifies the tool’s utility. Students, for instance, can turn a set of lecture notes into a study quiz automatically.

Smarter Source Discovery and Integration

NotebookLM’s reliance on user-provided sources has always set it apart from general-purpose chatbots like ChatGPT or Google Gemini. The tool only answers questions based on the documents you upload, which reduces hallucination risks but requires manual effort to gather relevant materials. The latest update eases this burden by allowing the AI to suggest and import web pages automatically. From the chat interface, users can ask NotebookLM to find sources that match their research topic. The system returns a “research report” listing potential web pages, and users can choose to import all or some of them. Once added, these sources become part of the notebook’s context for all future interactions.

This feature leverages Google’s search infrastructure, though the inner workings are not fully explained. It likely uses Gemini to understand the user’s intent and then retrieves relevant URLs. The ability to discover sources without leaving the tool could save researchers considerable time, especially when exploring unfamiliar topics. However, the quality of suggested sources will depend on the underlying search algorithms and the AI’s interpretation.

Rollout and Availability

These features are beginning to roll out today, but most users will not see them immediately. The deployment is tiered: AI Ultra subscribers gain first access, followed by Workspace business customers with AI Ultra Access and AI Expanded Access. Other Google account holders will receive the updates in the coming weeks or months. This staggered approach is typical for Google’s AI product launches, allowing the company to monitor performance and gather feedback before wider release.

NotebookLM’s evolution reflects Google’s broader strategy to embed AI into productivity tools while maintaining a focus on source-based reasoning. By combining a more powerful language model, code execution, rich output formats, and automated source discovery, the update aims to create an all-in-one research assistant. Whether it can compete with dedicated coding environments or specialized reporting tools remains to be seen, but the direction is clear: Google wants NotebookLM to be the starting point for any knowledge work, not just a note-taking app.

Source:Ars Technica News