Google NotebookLM adds Gemini 3.5, code execution via Antigravity, and 10+ export formats
Google upgraded NotebookLM to use Gemini 3.5 and its Antigravity coding tool, enabling code execution through a "secure cloud computer" with 100+ software skills. The system achieved a 78.2% win rate against the previous baseline in web research tasks and now exports to 10+ formats including PDF, XLSX, and PPTX.
Google upgraded NotebookLM today with Gemini 3.5 and its Antigravity coding tool, adding code execution capabilities and expanding export formats to more than 10 file types.
Code execution via Antigravity
Each NotebookLM notebook now runs what Google calls a "secure cloud computer" that executes code for research and analysis tasks. The system includes more than 100 curated software skills, according to Google.
The company claims the upgraded NotebookLM achieved an average 65% win rate across five core evaluation dimensions—a 15 percentage point margin above parity. In large document analysis, Google reports a 69.9% win rate, while web research and source discovery reached a 78.2% win rate against the prior baseline.
New export formats
NotebookLM now exports to these formats:
- Data visualizations: PNG, SVG
- Documents: PDF, DOCX, Markdown, text files
- Images (via Nano Banana): PNG, JPG, GIF
- Structured data: CSV, JSON
- Microsoft Office: XLSX (Excel), PPTX (PowerPoint)
Users can request detailed outputs like PDF reports with charts and tables or budget spreadsheets, then request edits after generation. Google says more formats are coming.
Enhanced research features
The chat interface now helps users build source repositories from "loose ideas and questions." NotebookLM can use Google Search to find and add web sources to notebooks, including primary sources in other languages.
Google provided three use cases: a data analyst combining multi-country datasets with code-based analysis, a program manager converting technical specs into simplified guides and slide decks, and a gym owner calculating media campaign ROI from raw sales data.
What this means
This positions NotebookLM as a competitor to Claude Projects and ChatGPT's canvas feature, with deeper integration into Google's productivity suite through native Office format exports. The code execution capability via Antigravity mirrors similar features in ChatGPT's Advanced Data Analysis and Claude's analysis tools, but Google's approach bundles it with web search and multi-format exports. The upgrade is rolling out today to Google AI Ultra and Workspace business customers with AI Ultra access, with other tiers coming "over time."
Related Articles
Google Rebrands NotebookLM to Gemini Notebook, Brings Gemini 3.5 and Antigravity to AI Pro
Google renamed NotebookLM to Gemini Notebook and announced that the Gemini 3.5 model with Antigravity code execution capability will roll out to AI Pro subscribers in the coming weeks. The research tool now has over 30 million users and 600,000+ organizations.
Google renames NotebookLM to Gemini Notebook, adds code execution for Ultra and Workspace users
Google is renaming its AI note-taking app NotebookLM to Gemini Notebook while keeping it as a standalone app. The company is also rolling out code execution capabilities to Google AI Ultra and Workspace business customers, with Pro user access coming in the following weeks.
Google Gemini Spark adds Workspace editing, gets 50% speed boost, expands to AI Ultra subscribers
Google has upgraded its Gemini Spark personal agent with the ability to edit shared Google Workspace documents, a 50% speed improvement, and smarter parallel source processing. The service is now available to Google AI Ultra subscribers in most regions, with AI Pro access planned for the near future.
NASA Runs Google's Gemma 3 Aboard Orbiting Satellite for On-Board Image Analysis
NASA's Jet Propulsion Laboratory ran a 4-bit quantized version of Google's Gemma 3 4B on board a Loft Orbital YAM-9 satellite, marking the first in-orbit demonstration of a vision-language model analyzing imagery from a satellite's own sensor. The unmodified open-weights model hit 88 percent accuracy on a 7,960-image benchmark and ran live capture tests over Toulouse, France and coastal Argentina.
Comments
Loading...