Google NotebookLM Upgrades to Gemini 3.5, Adds Web Search and Cloud Code Execution

Google has upgraded NotebookLM, its AI-powered note-taking application, with significant capability expansions centered on research workflows. According to The Verge AI, the redesigned platform now runs on Gemini 3.5, enabling what Google describes as more accurate and reliable responses, and introduces integrated web search plus cloud-based code execution to reduce friction in gathering and processing research materials.

Gemini 3.5 Integration and Accuracy Gains

The shift to Gemini 3.5 represents NotebookLM’s third model upgrade since its 2023 launch. According to The Verge AI, the new foundation model improves response quality, though Google has not published comparative benchmarks against the prior Gemini version. The upgrade applies across all core features: document Q&A, source interaction, and note synthesis remain unchanged in workflow, but the underlying reasoning and factual grounding benefit from Gemini 3.5’s training improvements.

Automated Source Discovery and Web Integration

NotebookLM previously required users to manually upload documents, PDFs, or YouTube video links before starting research. The update inverts this pattern: users can now pose a research question directly, and NotebookLM will invoke Google Search to locate relevant sources autonomously. According to The Verge AI, this builds on NotebookLM’s existing “discover” feature, which surfaces web resources. The new behavior essentially transforms NotebookLM from a document-interaction tool into a research assistant that can seed its own corpus.

Cloud Code Execution via Antigravity

NotebookLM notebooks now connect to what Google calls a “secure cloud computer,” powered by Google’s Antigravity agentic coding platform. According to The Verge AI, this enables NotebookLM to write, execute, and output code—expanding use cases to data analysis, visualization generation, and programmatic transformations that previously required export-to-external-tool workflows. The cloud infrastructure handles compute and storage, removing local resource constraints.

Export Format Expansion

The update adds multi-format output: PDF documents, image files (PNG, SVG, GIF, JPG—including Nano Banana-generated visuals), Excel and PowerPoint files, and CSV sheets. According to The Verge AI, this reduces friction for sharing research outputs across teams and publishing pipelines.

Rollout and Availability

Access is tiered: Google AI Ultra subscribers and Google Workspace customers receive the upgrade immediately. According to The Verge AI, Google intends to expand access to additional subscription tiers in the future, though no timeline or tier list was disclosed.

Why This Matters

NotebookLM’s shift from a closed-document interface to a web-connected research agent with code execution capability repositions it as a competitor to research platforms like Perplexity (which also integrates search) and to manual research pipelines (Jupyter notebooks + external search). For teams evaluating AI research tools, the cloud code execution and multi-format export reduce the need for context-switching to spreadsheet or visualization tools. However, the initial gating to AI Ultra and Workspace customers limits immediate impact to paying subscribers; broader adoption depends on the planned expansion timeline and pricing structure for lower-tier plans.