Verdict: The June 2026 update transforms NotebookLM from a retrieval tool into a fully autonomous research agent. By integrating Gemini 3.5 and the Antigravity IDE, the platform can now write code, run deep-web searches from zero sources, and generate professional-grade assets (PowerPoint, Excel, PDFs) in parallel—effectively acting as a secure cloud-based research assistant.
Last verified: 2026-06-23 · Rollout: Live for Google AI Ultra & Workspace Business (AI Ultra / AI Expanded Access) · Core Engine: Gemini 3.5 Flash + Antigravity.
- New trick: every notebook gets a secure cloud computer that writes and runs code, with 100+ built-in software skills.
- Source discovery: start from a blank notebook and a question; NotebookLM finds and cites web sources for you.
- Free tier still exists at $0 (100 notebooks, 50 sources/notebook, 50 chats/day) — but the agentic features are paid-tier first. Pricing and limits change often. Verify before choosing a tier.
What is the NotebookLM June 2026 Update?
On June 8, 2026, Google officially rolled out the "Agentic" upgrade to NotebookLM. This is not a minor UI tweak; it is a fundamental shift in how the system processes information. While previous versions required you to provide all source materials upfront (RAG architecture), the new version can autonomously build its own knowledge base and execute complex engineering or analytical tasks using a dedicated "secure cloud computer" attached to every notebook.
Three things changed at once: the model, the tool's hands, and how you start a project.
Key Features: Gemini 3.5 and Antigravity
The update is built on two primary pillars that increase the platform's reasoning and execution capabilities:
- Gemini 3.5 Integration: NotebookLM now defaults to the Gemini 3.5 Flash model. This upgrade brings visible reasoning steps, allowing users to inspect the "logic chain" the AI used to arrive at a conclusion.
- Antigravity IDE: Originally an internal Google coding tool, Antigravity is now embedded into NotebookLM. This allows the agent to write and execute code in real-time to solve data problems, create visualizations, or transform raw spreadsheets into structured analysis.
In Google's own side-by-side tests against the prior version, the upgraded system won more than 65% of the time across five core research tasks, scoring 69.9% on large-document analysis and 78.2% on advanced web research and source discovery. (Google internal evaluation, June 2026 — vendor claim, not independently reproduced.)
New Workflows: From Static Notes to Active Agents
The "agentization" of NotebookLM introduces three workflows that were previously impossible within the app:
1. Zero-Source "Deep Research"
Previously, NotebookLM was a "closed-loop" system. You uploaded PDFs, and it talked about them. Now, you can start with a blank page and a question. The system uses Google Search to identify high-quality primary sources, suggests them to you, and builds the source library dynamically as you chat. You stay in control: it proposes sources, you decide which go in.
2. Autonomous Cloud Computing
Each notebook now acts as a secure, sandboxed environment. The AI uses over 100 curated software skills to perform tasks that once required external tools. For example, you can ask the notebook to "Clean this CSV and generate a year-over-year growth chart," and it will execute the Python code internally to produce the file.
This is the biggest practical shift, and the reason to trust its arithmetic more than a plain chatbot's: when you ask for a calculation, NotebookLM writes real code, runs it, checks the result, and hands back a chart or table rather than guessing at a trend. The curated skills cover parsing messy spreadsheets, comparing differently formatted documents, pulling numbers out of PDFs and building visual reports.
3. Parallel Output Generation
The "Studio" feature now supports parallel execution. You can trigger a podcast overview, an infographic design, and a comprehensive research report simultaneously. The agents work in the background, allowing you to move to other notebooks while the assets are finalized.
NotebookLM Output Formats (June 2026)
The update significantly expanded the "Studio" capabilities. The system now supports high-fidelity exports across these formats:
| Format Type | Supported File Types | Best Used For |
|---|---|---|
| Documents | PDF, Word (DOCX), Markdown, plain text | Final reports, briefing docs, whitepapers |
| Spreadsheets | Microsoft Excel (.xlsx) | Budgets, models, cleaned data |
| Structured data | CSV, JSON | Database imports, dashboards, cleaned logs |
| Charts | PNG, SVG | Slide-ready visuals, diagrams |
| Images | PNG, JPG, GIF (via Nano Banana) | Infographics, custom cover art |
| Multimedia | Audio Overviews (podcast), Slides (PPTX) | Stakeholder presentations, educational content |
You can also edit a file after generation by giving follow-up instructions the way you would to a colleague.
How much does NotebookLM cost in 2026?
NotebookLM is not sold on its own; it is bundled inside Google's AI plans. Google's official support table breaks the consumer tiers down like this:
| Plan | Price (USD/month) | Notebooks | Sources/notebook | Chats/day | Audio Overviews/day | Deep Research |
|---|---|---|---|---|---|---|
| Standard | Free | 100 | 50 | 50 | 3 | 10/month |
| Plus (Google AI Plus) | $7.99 | 200 | 100 | 200 | 6 | 3/day |
| Pro (Google AI Pro) | $19.99 | 500 | 300 | 500 | 20 | 20/day |
| Ultra 20 TB (Google AI Ultra) | $99.99 | 500 | 500 | 2,500 | 100 | 75/day |
| Ultra 30 TB (Google AI Ultra) | $200.00 | 500 | 600 | 5,000 | 200 | 200/day |
These limits were last checked against Google's help page on 2026-06-17. The agentic features — secure cloud computer, web source discovery, expanded exports — reached Google AI Ultra subscribers and Workspace business customers with AI Ultra Access or AI Expanded Access first.
Is it worth paying for?
For most solo users and students the free tier still covers note-taking and basic Q&A. Paid tiers start to make sense when one of these is true:
- You hit the 50-source or 50-chat daily cap regularly.
- You want messy spreadsheets turned into clean charts or reports without touching Excel or Python.
- You run market or competitor research and want NotebookLM to find and cite sources.
- You create a lot of Audio Overviews, Video Overviews or slide decks.
The $19.99/month Pro tier is the practical sweet spot for a small-business power user: 500 notebooks, 300 sources each, 500 chats/day, 20 Deep Research runs/day. The Ultra tiers are enterprise-scale; most small teams do not need them.
How to use the new NotebookLM: a 5-step workflow
- Start with the outcome. Decide whether you want a report, chart, slide deck, spreadsheet or cleaned data file before you type the prompt.
- Give it sources or a question. Either upload your files or ask it to find web sources on a topic. Review the sources it picks.
- Ask for a specific artifact. For example: "Build an Excel file showing monthly revenue by product, with a chart and a short written summary."
- Check the work. Open the exported file and spot-check a few numbers or citations. The tool runs code, but you are still the editor.
- Iterate in plain English. "Make the chart a bar graph," "Add a column for gross margin," "Shorten the summary to three bullets."
What this means for you
For small businesses and individual researchers, this update eliminates the "capture gap." You no longer need to spend hours gathering and formatting data before you can begin analysis. By chaining NotebookLM with other high-context models, you can build a self-sustaining research engine that monitors your industry and delivers cited, verified briefs every morning.
This shift toward autonomous AI agent orchestration marks the end of the "chatbot" era and the beginning of the "worker" era, where the tool handles the grind of data processing while you focus on the verdict.
Three high-value first moves:
- Clean up your messy data. Upload sales, refund or lead-source spreadsheets and ask for the one metric that actually matters — "Which product is worth pushing harder?" The cloud computer runs the maths instead of hallucinating a trend.
- Turn dull docs into usable guides. Feed it a long policy manual, onboarding doc or technical spec and ask for a one-page team guide plus a slide deck.
- Start research from a question, not a folder. Open a blank notebook, ask about a market or competitor, and let it find and cite credible sources — then edit the source list before trusting the answer.
By role
- Founders and operators: turn raw business data into a decision-ready report. Code execution means you can trust the calculations more than a plain chatbot reply.
- Marketers: feed it a batch of customer reviews, support tickets or competitor pages and ask for a content brief, FAQ or a single image that summarises the main finding.
- Consultants and coaches: drop in call transcripts or training notes and ask for a one-page client guide or onboarding deck. For a similar workflow with a different tool, see our Claude progression guide.
- Solo builders: pair it with your other assistants. If you already use Gemini, our Gemini 2026 guide shows where NotebookLM fits in the same Google stack.
The tool's real edge is speed. One person doing five jobs can hand the grunt work of formatting, calculating and sourcing to the agent, and spend their own time on the decision.
Related reading
FAQ
Q: Do I need a paid subscription for the June 2026 update? A: The core NotebookLM features remain free. However, the advanced agentic features (Cloud Computer, Antigravity-powered skills, and Parallel Studio outputs) are currently rolling out to Google AI Ultra and Workspace Business (AI Ultra/Expanded) accounts first. Broader availability is planned but Google has not set a date.
Q: What is the "secure cloud computer"? A: An isolated, per-notebook sandbox where NotebookLM writes and runs code to analyse your data. Google says it is locked to that notebook, so your files and queries are not mixed with other users' data while the code runs.
Q: Can I still use my own sources exclusively? A: Yes. Privacy remains a core feature. The "Deep Research" mode is an optional add-on. Your private notebooks are still secure and are not used to train global Google models unless explicitly shared.
Q: Does it actually write code? A: Yes. It uses the Antigravity IDE to write and execute Python code within a secure cloud sandbox. This is how it handles complex Excel calculations and professional chart generation.
Q: What files can it export now? A: PDF, DOCX, Markdown, plain text, XLSX, CSV, JSON, PNG, SVG, JPG, GIF and PPTX, plus Nano Banana image generation inside the notebook.
Q: How do the citations work now? A: Citations are more granular. When the agent uses web-searched sources or your own PDFs, every claim is backed by a clickable number that takes you directly to the relevant snippet in the source document.
Q: Can it generate images for presentations? A: Yes, it integrates with the Nano Banana model to generate relevant, high-quality visuals for slide decks and infographics directly within the Studio.
Q: How is this different from ChatGPT or Claude with file uploads? A: NotebookLM is built around source-grounded answers with citations, and now around code execution in a secure sandbox. The trade-off is that it is less of a general chatbot and more of a structured research workspace.
Q: Is my data safe in NotebookLM? A: For individual accounts, Google's help page says data is protected and not used to train NotebookLM unless you submit feedback. For Workspace and Education users, uploads, queries and responses are not reviewed by humans and are not used to train AI models. Review your own organisation's policy before uploading sensitive customer or financial information.
Every claim here is traced to a primary source, dated, and listed under Sources. Research and drafting are AI-assisted; editing, verification and publication are human decisions, and a person is accountable for what appears on this page. How we work →
Discussion
0 comments