Google recently launched a series of functional upgrades for its AI note-taking and research tool NotebookLM, focusing on improving reasoning capabilities in complex research scenarios, and adding a variety of result output formats, which are open to paying users and Workspace enterprise customers around the world.
According to reports, after this upgrade, NotebookLM’s underlying model has been fully switched to Gemini 3.5 and combined with Antigravity technology to improve the accuracy and reliability of results. In response to the "opaque thinking process" issue that users have long been concerned about, NotebookLM now displays more detailed "reasoning steps" in the chat interface, allowing users to see how the system extracts information from the data and draws conclusions.
In its internal evaluation, Google compared the new version of NotebookLM with the old version, covering five core dimensions including accuracy and quality, multi-language support, large document analysis, document creation, and advanced research. The results show that the new version's average "win rate" on these metrics exceeds 65%, which is 15 percentage points higher than the baseline level, showing a significant performance improvement. Among them, in terms of large-volume document analysis, the winning rate of the new version of NotebookLM reached 69.9%; in terms of advanced web page research and information source discovery, its winning rate reached 78.2%, which was described by Google as having achieved "excellent performance."
In terms of output methods, NotebookLM has also been greatly expanded, supporting more types of file formats, making it easier for users to use AI-generated content directly for visual display, structured data analysis, or office document production. Specifically, data visualization results can be exported to PNG and SVG; document content can be exported to PDF, DOCX, Markdown, and plain text files; image content can be exported to PNG, JPG, and GIF; structured data supports JSON and CSV; and direct output support for Microsoft Excel (XLSX) and PowerPoint (PPTX) formats has also been added. Users can still modify the results after generating them, and download the corresponding files through the "Studio Panel" in the interface.
Google said that these new capabilities are open to the world. Users can not only issue commands in one language and generate results in another language, but also rely on NotebookLM to build a database during the project startup phase. Even if a complete list of information is not prepared, users can talk to keywords with only some preliminary ideas, and NotebookLM will actively help discover and organize relevant information sources during the chat process. For example, it can help find primary sources in other languages to gain a different perspective, or to expand on an author's related works and research.
These updates are currently available to users with Google AI Ultra subscriptions and Workspace enterprise customers with AI Ultra access. Google also stated that it plans to gradually promote the above functions to a wider range of user groups in the future, but has not yet announced a specific timetable.