Google's NotebookLM Can Now Run Code, Find Its Own Sources, and Generate Presentations

A new brain under the hood: Gemini 3.5 and Antigravity

The most fundamental change is the switch to Gemini 3.5 as NotebookLM's default reasoning engine . Google positions this as a leap in “accuracy and reliability,” with “better visibility into the thinking process” . Early benchmarks suggest the gains are real. In head-to-head evaluations, the upgraded NotebookLM scored an average “win rate” above 65%, roughly 15 percentage points higher than its predecessor. Performance spiked further in specialized tasks: 69.9% for deep document analysis, and 78.2% for advanced web research and source discovery .

Antigravity, Google's agent-first IDE, is what makes the new autonomous behaviors possible . It equips each notebook with a secure, sandboxed cloud computer, effectively turning the notebook into an environment where AI can write and execute code to complete a task . This isn't a sidebar code interpreter; it's a foundational shift that allows NotebookLM to act, not just answer.

What NotebookLM can now do on its own

The Antigravity integration unlocks two major capabilities that move NotebookLM beyond a simple chat interface.

Autonomous code execution means the tool can now perform complex data analysis directly. Instead of describing what your spreadsheet should look like, you upload raw data and NotebookLM writes and runs the necessary code in its secure environment to clean, analyze, and visualize it . This removes the manual step of exporting data to a separate analytics tool.

Chat-based source discovery solves the blank-page problem. You can now start a project with a loose concept, and NotebookLM will help you build a knowledge base from the web. It uses its own research skills and Google Search to suggest and pull in relevant sources, reducing the upfront work of manually curating every document . Google notes this can be especially powerful for finding primary sources in other languages or related material from unexpected angles .

From text answers to finished deliverables

The update dramatically expands what NotebookLM can produce. It's no longer confined to text summaries and Q&A. Users can now generate and export in 10 distinct file formats, covering the full spectrum of modern office work :

• Documents: PDF, DOCX, Markdown, and plain text
• Spreadsheets: Microsoft Excel (XLSX) and CSV
• Presentations: Microsoft PowerPoint (PPTX)
• Visualizations: Charts exported as PNG and SVG
• Structured data: JSON
• Images: PNG, JPG, and GIF

These outputs are generated from your sources and can be modified before download through a new “studio panel” interface . As one tester noted, NotebookLM is evolving from a smart research assistant into a tool that can create a wide range of finished, downloadable deliverables .

A word on access and the missing free tier

Google's official blog makes the rollout terms clear: the update is “available globally today for users with Google AI Ultra and specific Workspace business accounts” . A Chinese-language tech outlet corroborated this, describing the feature set as open to “paid users and Workspace enterprise customers worldwide” .

What none of the launch-day coverage clarifies is whether or when standard free users will get access. Given the computational heft required for the secure cloud computing environment and agentic workflows, it's plausible Google is deliberately gating the most resource-intensive features behind its premium subscription. Until Google updates its official NotebookLM landing page or support documents, free-tier users should assume they'll see a narrower feature set.