What NotebookLM’s New Agentic Research Model Actually Is
NotebookLM’s new agentic research model is an AI research tool upgrade that combines Gemini 3.5, code execution, and multi-format output generation so the system can plan, run, and complete complex research workflows with far less manual prompting while keeping users in control of sources and citations. Originally launched as an experimental product from Google Labs, NotebookLM has grown into a collaborative research and knowledge platform used by millions of people and organizations to organize documents and uncover hidden links. The latest release adds a more thoughtful chat experience that shows more of the AI’s reasoning while it works through large document collections or web sources. Instead of acting as a passive question-answer bot, NotebookLM now behaves more like an active research assistant that can help define a project, gather material, analyze it, and package results into ready-to-share files.
Agentic capabilities powered by Gemini 3.5 and Antigravity
At the core of the upgrade are new NotebookLM agentic capabilities driven by Gemini 3.5 and Google’s Antigravity technology. The chat experience has been rebuilt so the system can autonomously break down complex questions, propose next steps, and carry out multi-step research tasks instead of waiting for tightly scripted prompts. Google reports that in internal tests, the upgraded NotebookLM achieved an average win rate of over 65% against the prior system, with a 69.9% win rate in large-document analysis and 78.2% in advanced web research and source discovery. These gains show up in practice when you ask NotebookLM to survey a field, compare several reports, or build a structured summary from messy source collections. The tool now also helps users start from a blank page, turning a vague idea or initial question into a curated source library inside the chat while preserving clear attribution.
Code execution and data analysis inside your notebook
One of the most consequential changes is code execution analysis inside each notebook. Every NotebookLM workspace now ships with a secure cloud computer that can write and run code, so the AI can move beyond summarizing text and into direct computation, data cleaning, and visualization. Google says this environment exposes more than 100 curated software skills, giving users access to specialized tools without leaving the research flow. That means NotebookLM can, for example, ingest multiple datasets, reconcile formats, compute metrics, and plot charts, then feed the results back into the conversation for further advanced reasoning. Data analysts can explore complex CSV files, researchers can test hypotheses with quick scripts, and small business owners can compare sales performance against marketing activity without juggling separate coding and spreadsheet tools.
Advanced reasoning features that connect ideas and sources
The agentic upgrade is matched by advanced reasoning features that change how NotebookLM connects disparate information. Running on Gemini 3.5, the system is better at tracking context across long chats and large collections of documents, linking themes, arguments, and data points that might otherwise be missed. Internal evaluations highlight this improvement in large-document analysis and source discovery, which matters when you are working through dense research papers, technical specifications, or evolving project notes. NotebookLM can now combine web research with uploaded files, identify primary sources in multiple languages, and surface how a claim is grounded in specific passages. The reasoning process is more transparent too, giving users clearer views of how answers were derived, which helps with fact-checking and academic or professional rigor while still keeping the human in charge of final judgments.
From raw research to polished outputs and automated workflows
The final piece of NotebookLM’s upgrade is output expansion, which turns agentic research into practical deliverables and research workflow automation. Once analysis is complete, NotebookLM can generate polished outputs directly from your sources: PDF reports with charts and tables, DOCX documents, Markdown and plain text summaries, structured datasets in CSV and JSON, Excel spreadsheets, and PowerPoint presentations. It can also export images in PNG, JPG, GIF, and SVG formats, and these generated assets remain editable after creation, so teams can refine them in familiar tools. According to Google, users are no longer constrained to starting with pre-assembled document sets; they can begin with a question and end with shareable files that slot into existing workflows. This spans student projects, internal company documentation, client-facing decks, and cross-language collaborations that need consistent, well-structured research outputs.
