What NotebookLM’s Gemini 3.5 Upgrade Changes
NotebookLM is a cloud-based notebook and AI research tool that combines source-grounded analysis, document organisation, and now code execution to help researchers move from raw material to finished outputs in a single workspace. With the move to NotebookLM Gemini 3.5, Google is reframing the product from an experimental helper into a more serious research environment built around reasoning and traceable results. The new model brings stronger multi-step reasoning, better handling of long documents, and more transparent answer construction. Google’s own evaluations report that the upgraded NotebookLM achieves an average win rate of more than 65% over the previous version across five core dimensions, including a 69.9% win rate for large document analysis and 78.2% for advanced web research and source discovery. For practitioners, the promise is less about abstract scores and more about fewer dead ends in complex, document-heavy projects.
Smarter AI Reasoning for Deep Research Workflows
The move to NotebookLM Gemini 3.5 is aimed at research scenarios where questions evolve over time and sources pile up. Instead of requiring users to arrive with fully prepared materials, NotebookLM can now start from an open-ended question, use Google Search to suggest web sources, and let researchers select what becomes part of a notebook. This supports exploratory work such as literature reviews, competitive scans, or multi-language source hunting. Google says the upgraded system achieved a 78.2% win rate in advanced web research and source discovery when compared with the prior NotebookLM baseline. For complex document interpretation, more than 100 curated software skills sit behind the scenes, giving the model ways to inspect sources, run checks, and generate structured artifacts. The visible answer steps and Deep Research features help users see how conclusions connect back to the original material, which matters for academic and professional review.
Antigravity Code Execution: A Built-In Cloud Computer
Antigravity code execution adds a secure cloud computer to every notebook, turning NotebookLM into more than a text-only assistant. Within a single project, researchers can move from reading to analysis by writing and running code directly in the interface, without exporting data to a separate coding tool. This is particularly useful for statistical checks, custom parsing of messy documents, or building repeatable analysis pipelines. Antigravity 2.0 provides an agentic coding layer that can decide when to call tools, inspect intermediate results, and refine outputs. A single research question can now chain together source selection, programmatic analysis, and formatted reporting. NotebookLM remains source-grounded: the code execution environment is aligned with the notebook’s materials rather than a general-purpose IDE. That design keeps the product focused on AI research tools that produce traceable, document-linked results instead of turning into a wide-open coding sandbox.
From Notes to Reports: Expanded Outputs and Exports
NotebookLM now behaves more like a full publishing pipeline than a note-taking app. Once analysis is done, users can export their work in multiple formats, turning source-backed answers into reports, data files, or presentations. The upgraded system can generate PDF reports with charts and tables, Word documents, Markdown and plain text, as well as CSV, JSON, and Excel files. It can also create images and PowerPoint decks, which means a single workflow can run from raw documents to executive-ready slides without manual copying. Outputs are grounded in user sources and can be further refined through a studio panel that lets researchers adjust instructions and regenerate sections. This helps teams standardise deliverables—such as consistent report templates or recurring briefing decks—while keeping a clear link to the evidence that fed each section. For recurring analysis, the same notebook can act as a living knowledge base and export engine.
Tiered Access and the Road to a Unified Research Platform
Access to NotebookLM Gemini 3.5 and Antigravity code execution is structured as an early, tiered rollout. Google AI Ultra subscribers and eligible Workspace customers are first in line, while wider availability is planned but not yet fully open. This strategy keeps the initial deployment focused on users most likely to stress-test advanced features: research-heavy teams, technical professionals, and students with complex workflows. NotebookLM’s evolution also reflects broader competition among AI research tools such as Claude, Elicit, Perplexity Spaces, and enterprise-focused notebook products. Google is trying to combine trusted retrieval, long-lived notebooks, and cloud-based code execution into one environment rather than scattering tasks across chatbots, IDEs, and document editors. If the real-world performance matches Google’s internal figures, NotebookLM could move from a helpful companion to a primary workspace for source discovery, structured reasoning, and reproducible, cloud-based notebook analysis.
