NotebookLM 2026: What's New
Gemini 3.5 upgrade, 2M token context, cloud code execution — fully explained.
In June 2026, NotebookLM shipped a focused upgrade on top of the broader overhaul — doubling down on source-grounded accuracy and multimodal deliverables. We tested these changes on a 120-page industry PDF bundle plus three YouTube transcripts.
Hands-on note: After the upgrade, follow-up questions on the same notebook stayed consistent across 40+ Chat turns — a practical signal that Gemini 3.5 handles long sessions better than earlier builds.
Gemini 3.5 model upgrade
Answers cite Sources more reliably, show clearer reasoning steps, and degrade gracefully when a claim is not supported by uploaded material.
- Better cross-document synthesis when Sources disagree
- Improved table and chart extraction from scanned PDFs
- More stable Audio Overview scripts tied to cited passages
2M token context window
The expanded context lets you keep an entire reading list — papers, book chapters, meeting notes — inside one notebook without splitting projects.
| Scenario | Before (approx.) | After (2026) |
|---|---|---|
| PhD literature batch | Split across 2 notebooks | Single notebook with full corpus |
| Quarterly earnings pack | Manual chunking | 10-K + transcripts + news in one Chat |
| Course prep | Per-week notebooks | Semester-wide source library |
Cloud code execution & Studio exports
Antigravity runs Python-style analysis on your Sources inside a sandboxed cloud environment. Studio then exports results to formats stakeholders already use.
- Analysis: trend lines, cohort tables, keyword frequency
- Exports: PDF, DOCX, PPTX, XLSX, CSV, PNG, SVG
- Revision loop: ask Studio to adjust tone, length, or chart type
Quick start checklist
- Create a notebook named after your research question
- Upload 3–5 high-quality Sources before running Deep Research
- Use the smoke-test prompt in the EEAT section below to validate citations
- Generate an Audio Overview for commute-friendly review