What actually changed: From notes to autonomous research

The original NotebookLM worked as a smart assistant that could analyze your uploaded documents and answer questions. The new version takes this concept several levels higher.

The foundation is the transition to Gemini 3.5 Flash — Google's fastest model — and integration with Antigravity, Google's developer platform for agentic coding. Thanks to this, each notebook gets its own virtual computer that can run code directly in a secured environment. The system features more than 100 ready-made software skills, from statistical analysis through creating visualizations to advanced data work.

Google also published concrete numbers from internal testing. The new system beat the previous version in 65% of cases across five main categories. In large document analysis, it achieved a success rate of 69.9% and in advanced web research even 78.2%. These aren't cosmetic improvements — it's a leap in capabilities that makes NotebookLM one of the most powerful AI research tools on the market.

From idea to finished output: New export formats

One of the biggest practical innovations is the expansion of output formats. Whereas previously you got mostly a text summary from NotebookLM, now you can directly generate:

PDF reports with charts and tables, Excel spreadsheets (XLSX format), PowerPoint presentations, but also data visualizations in PNG and SVG formats, structured data in CSV and JSON, or images generated using Nano Banana. For anyone working with data or preparing materials for a team, this is a massive time saver — you can handle analysis, visualization, and presentation of results in a single tool.

Moreover, Google promises that outputs can be further edited after generation and plans to add more formats in the future. To give you an idea: a data analyst can upload raw data from several countries with inconsistent formatting, have NotebookLM look up additional context on the web, run code for precise analysis, and generate a PDF report with charts — all within a single notebook.

Research from scratch: You no longer need your own sources

A fundamental change is the so-called "zero-source" mode. Previously, you had to enter NotebookLM with your own documents — without uploaded PDFs or text files, the tool was useless. Now you can just enter a free thought or question and NotebookLM will search the web itself via Google Search, find relevant and quality sources, and add them to your notebook.

For example, you can start with the question "what is the current situation with AI regulation in the EU" and let the tool find primary sources, legislative documents, and expert analyses on its own. This opens the door to a completely new style of work: instead of hours of googling, you let AI build your research base, and you focus on what matters — analysis and interpretation.

Importantly, you remain in control over which sources are added to the notebook. All information continues to be clearly attributed and the work remains anchored in trustworthy sources.

Price and availability in Czechia

The new NotebookLM is available globally on the web from June 8, 2026 for all Google AI Ultra users and for Google Workspace enterprise customers with AI Ultra Access or AI Expanded Access subscriptions.

Google AI Ultra is available in Czechia in several tiers. The basic Ultra 5× plan costs approximately 1,200 CZK per month and includes 20 TB of cloud storage, extended access to Gemini 3.1 Pro models, Deep Research, and also YouTube Premium. The top Ultra 20× plan costs roughly 2,400 CZK per month with 30 TB of storage and the highest limits for all AI features.

For companies and institutions, a standalone Gemini Enterprise subscription is available, which can be added to an existing Google Workspace account — the price depends on the number of users and specific configuration. In the context of the European market, it is significant that Google AI Ultra is available in more than 150 countries including the entire EU and the Czech Republic.

How it stacks up against the competition

NotebookLM isn't alone in the market. ChatGPT from OpenAI offers similar features in the form of uploaded file analysis and Deep Research (available in the Pro plan for $200/month, roughly 4,800 CZK). Claude from Anthropic excels at working with long documents and in the Pro version ($20/month, roughly 480 CZK) can analyze extensive texts. Perplexity scores points in research with automatic source discovery.

But what sets NotebookLM apart is precisely the combination of everything in one: code execution, formatted output generation, and automatic source discovery. None of the main competitors currently offer the ability to have AI write and run code on your data while also generating a PowerPoint presentation from it. For students, academics, and data analysts, this is unique value.

On the other hand, being locked behind a relatively expensive Google AI Ultra subscription (from ~1,200 CZK/month) means that an average student on a coffee budget probably won't afford the new version. The cheaper Google AI Pro plan (~720 CZK/month) offers only more limited access to NotebookLM features. It's a strategy that Google clearly targets at professional users and companies — unlike the original NotebookLM, which was available for free.

What this means for Czech users

For Czech university students, scientists, and analysts, the new NotebookLM brings several concrete advantages. Czech is supported — you can upload Czech documents, ask questions in Czech, and receive answers in Czech (although English still provides the best results, especially for technical topics).

From the perspective of the EU AI Act, it is interesting that Google processes data within its global infrastructure — for companies working with sensitive data, it is therefore crucial to verify whether using NotebookLM does not interfere with obligations arising from European regulation, especially concerning personal data or trade secrets.

NotebookLM also fits into a broader trend where AI tools are shifting from simple chatbots to full-fledged work environments. It's no longer just about asking AI — it's about letting it independently work with your data, program, analyze, and produce finished outputs.