What NotebookLM’s Gemini 3.5 Upgrade Actually Is
NotebookLM’s Gemini 3.5 upgrade is an AI-powered research environment that combines source-grounded chat, automated source discovery, and secure cloud-based code execution to handle complex, multi-step research tasks from first question to finished output. Built on Google’s Gemini 3.5 model and Antigravity tool layer, the new NotebookLM moves beyond static note-taking into an active research partner for students, researchers, and workplace teams. Users can now start from a loose question, pull in PDFs, web pages, Docs, Slides, YouTube videos, and other media, and let the system propose relevant sources, analyse them, run calculations, and turn findings into exportable reports or presentations. According to Google’s internal benchmarks, the updated NotebookLM “achieved an average win rate of more than 65 percent across its top five core evaluation dimensions” compared with the previous version.
Deeper Reasoning and Transparent Answer Steps
Running on Gemini 3.5 gives NotebookLM more advanced reasoning for large document analysis and complex queries, while Antigravity adds a tool layer that can inspect sources and structure multi-step tasks. In chat, users see expanded reasoning steps, making it easier to judge how answers were produced and to trace claims back to cited material. Google reports a 69.9 percent win rate in large document analysis and a 78.2 percent win rate in advanced web research and source discovery against the prior baseline, suggesting clearer gains in both depth and breadth of research. Source control remains central: users choose what goes into each notebook, and answers stay grounded in those materials with visible attributions. This focus on reasoning visibility and source tracing aims to make AI research tools more reliable for academic writing, policy work, or internal business analysis, where evidence trails matter.
Antigravity’s Code Execution Turns Notebooks Into Research Labs
The most striking change for researchers and technical users is Antigravity’s secure cloud computer inside every notebook, effectively turning NotebookLM into a code execution notebook for analysis. Instead of copying data into a separate IDE or notebook, users can ask NotebookLM to write and run code directly against their sources to perform calculations, simulations, or data cleaning. Google says the system includes more than 100 curated software skills, expanding what the AI can do with text, tables, and structured data. A single question can now trigger source discovery, data extraction, code-based analysis, and a summarised explanation in one continuous flow. For data-heavy fields such as economics, engineering, or social science, this agentic research automation trims the friction between reading, coding, and explaining, while keeping everything tied back to the original source set in the notebook.
From Question to Report: Agentic Research and Source Discovery
NotebookLM no longer expects users to arrive with a neatly curated reading list. The Gemini 3.5 upgrade adds Google Search–based source finding, so users can begin with questions and let the system suggest relevant web pages, primary materials, or related work by the same author. Selected items then become part of the grounded source set for ongoing analysis. This is where agentic research automation shows its value: NotebookLM can chain tasks such as searching for sources, filtering them, extracting key passages, and synthesising answers while keeping the user in control of what is included. The tool’s Deep Research features and reasoning visibility allow users to review intermediate steps, refine the query, or add new materials midstream. For students building literature reviews and professionals scanning unfamiliar domains, this cuts setup time and helps avoid missing important but non-obvious sources.
Downloadable Outputs and a Tiered Rollout
Once the analysis is done, NotebookLM can turn insights into a wide range of downloadable artifacts, trimming time spent on documentation and formatting. Users can generate reports, charts, documents, spreadsheets, slide decks, images, structured data files, and more from within the studio panel. Supported formats include PDF, DOCX, XLSX, PPTX, CSV, JSON, markdown, PNG, SVG, JPG, GIF, and plain text, making it easier to plug AI-generated work into academic submissions or workplace tools. Access, however, is limited for now: the Gemini 3.5 and Antigravity upgrade is rolling out on the web to Google AI Ultra subscribers and Workspace customers with AI Expanded access, with plans to broaden availability later. This tiered rollout means early benefits go to enterprise and premium users, while others may need to wait before these new AI research tools reach their everyday study or project workflows.
