What NotebookLM’s Gemini 3.5 Upgrade Changes
NotebookLM is an AI research assistant and note‑taking tool that helps users collect sources, organize knowledge, analyze complex materials, and generate structured outputs, now upgraded with Gemini 3.5 and cloud-based compute to support deeper reasoning, autonomous workflows, and multi-format exports for more efficient research projects across documents and the web. With Google’s Gemini 3.5 and the Antigravity platform, NotebookLM now aims for more accurate, reliable answers and clearer reasoning steps, turning it from a summarizer into a partner for advanced research analysis. Google’s own internal benchmarks report an average win rate above 65% across five core evaluation categories compared to the previous version, including 69.9% on large document analysis and 78.2% on advanced web research and source discovery. For researchers, students, analysts, and writers, this means NotebookLM Gemini 3.5 can support more demanding multi-source tasks with less manual digging and cross-checking.
From Summaries to Agentic AI Research Assistant
Earlier versions of NotebookLM depended on users arriving with a well-prepared pile of PDFs, notes, and links. The new release changes the starting point: you can drop a rough idea or question into the chat between the Sources and Studio panels, and NotebookLM will suggest building a source library around it. It uses Google Search to find high-quality, relevant material, surfaces options like primary sources or related works, and lets you approve or discard each source so you stay in control of citations. This marks a shift toward agentic capabilities, where the AI research assistant takes on parts of the research workflow autonomously rather than waiting for every manual input. It is especially helpful for multi-source analysis, since the system can both discover new material and trace where each fact came from, reducing the risk of losing context across long research sessions.
Cloud-Based Code Execution and Data-Centric Workflows
One of the most important changes for power users is secure cloud-based code execution inside every notebook. Instead of exporting tables or copying data into separate tools, NotebookLM can now write and run code on a dedicated cloud computer to perform deeper quantitative or exploratory analysis. Google says the upgraded environment includes more than 100 software tools and skills, covering tasks like data cleaning, statistical checks, and generating visualizations from your sources or uploaded datasets. This turns NotebookLM into a practical workspace for code execution research, where you can interrogate reports, run calculations, and immediately ask follow-up questions about the results. For analysts and technical researchers, it cuts down on context switching between notebooks, IDEs, and chat tools, and makes it easier to maintain an auditable research trail tied directly to your original sources and generated code.
PDF Export Notes and Rich Output Formats
NotebookLM’s export system is now built for publishing, not only drafts. From the Studio panel, you can send outputs into widely used formats: PDF reports with charts and tables, DOCX, Markdown, TXT, and structured data formats like CSV and JSON. Visual assets such as charts and other data visualizations can be exported as PNG or SVG, while Nano Banana images support PNG, JPG, and GIF. Spreadsheet and presentation workflows are covered with XLSX and PPTX export. You can give NotebookLM detailed export instructions, then review and request changes after a file is generated, rather than re-running everything from scratch. For teams, this makes PDF export notes and other deliverables easier to share, archive, and reuse across tools. Research outputs can move straight from NotebookLM into slide decks, dashboards, or documentation systems with fewer manual formatting passes.
What This Means for Future Research Workflows
Combined, Gemini 3.5 reasoning, cloud code execution, and rich export options move NotebookLM from an AI note-taking tool toward a full research environment. Complex, multi-step projects that once required separate tools for literature review, coding, charting, and report writing can now happen in a single notebook, with the AI coordinating many of the steps. The system’s improved multilingual behavior and source-building from a blank page help more people start projects faster, even when dealing with foreign-language material or unfamiliar topics. For now, the new features are rolling out to Google AI Ultra and Workspace business customers with AI Ultra access, but the underlying pattern is clear: AI research assistants are evolving into agentic systems that can propose sources, run analyses, and package outputs with far less user intervention, while still leaving humans in control of direction and final judgment.
