Which Gemini model does NotebookLM actually use? The full version history, verified

The full version history

Google has never published a single changelog page tracking NotebookLM/Gemini Notebook's underlying model version, so this history is reconstructed from Google's own announcement posts where they exist, and flagged as press-sourced or unconfirmed where a primary citation doesn't exist. If you've read a stale "NotebookLM runs on Gemini 1.5 Pro" claim somewhere, this table shows exactly how many upgrades have happened since that was last true.

Date              Model                          Confirmed by
--------------------------------------------------------------------------------------------
May 2023 launch   Gemini Pro / PaLM 2-era stack   General launch coverage; no single primary source
                                                    pins the exact split (launched as "Project Tailwind")
~2024             Gemini 1.5 Pro                  Press coverage; exact month unconfirmed
2024-early 2025   Gemini 2.0 Flash                Press coverage; exact date unconfirmed
May 2025          Gemini 2.5 Flash                Tied to the Android/iOS mobile app launch
October 2025      (context window expanded         Same 2.5 Flash model, larger context ceiling
                   toward Gemini's 1M-token cap)
19 Dec 2025       Gemini 3                        NotebookLM's own official X/Twitter post:
                                                    "officially built on Gemini 3!" (Pro vs Flash
                                                    variant not stated); see 9to5google's writeup
8 Jun 2026        Gemini 3.5 + "Antigravity"      Press coverage (ppc.land, 9to5google); Ultra +
                  agentic framework, cloud          enterprise users at launch, free tier
                  computer, 11 export formats       explicitly excluded
16 Jul 2026       (renamed to Gemini Notebook;     Google's official blog.google rename post, which
                   3.5 rollout continuing)           never names "Gemini 3.5" explicitly in its own text
~5 Aug 2026       Pro tier rollout completed        Press coverage only; not confirmed on a primary
                                                    Google page

Sources checked directly: blog.google's rename announcement, NotebookLM's own X
account, Google's Gemini Notebook support FAQ. Where a row says "press coverage,"
no equivalent primary-source Google page could be found stating that date.

Two of these dates are worth reading in the original coverage rather than taking on faith. 9to5Google's report on the Gemini 3 upgrade is the clearest secondary confirmation of the December 2025 switch, built directly on NotebookLM's own social post rather than a Google blog entry, since Google never published a dedicated announcement for that specific upgrade. The June 2026 jump to Gemini 3.5 is documented in more detail by ppc.land's writeup and 9to5Google's own coverage of the code-execution and 11-format upgrade, both of which specify the Ultra-and-enterprise-first rollout that Google's own rename post later referenced without repeating the model name.

Why Google doesn't name the model in its own copy

This isn't an oversight. Across Google's official Gemini Notebook rename post and its support FAQ pages, the consistent pattern is describing the product in terms of capability and subscription tier, not model version number. That's a deliberate framing choice consumer-product teams commonly make so that a future silent model swap doesn't require rewriting user-facing documentation, but it does mean that every specific model name in circulation, Gemini 3, Gemini 3.5, comes from press coverage and NotebookLM's own social posts rather than from an evergreen Google page you can check today and trust to stay current.

Does the model differ by tier or feature?

By subscription tier: the clearest documented split is that the June 2026 Gemini 3.5 upgrade launched for Ultra subscribers and enterprise users first, reached general Pro availability by early August 2026 per press coverage, and explicitly excluded the free tier at launch. No source confirms whether the free tier has since been brought up to the same model; until Google states that directly, the honest answer is unconfirmed, not "probably yes."

By feature within the product: no source we could find documents a distinct chat model for Deep Research specifically versus ordinary chat inside Gemini Notebook. The one confirmed exception is Video Overviews, whose visual styles and animated illustrations come from Nano Banana, Google's dedicated image-generation model, rather than whatever text model is currently answering chat questions. That's worth knowing if a chat-model upgrade doesn't seem to change how Video Overviews look: it's a genuinely different model doing that specific job. Audio Overviews are an older, separate feature, and no source documents them running on Nano Banana specifically; treat any claim connecting the two as unconfirmed.

By deployment (Enterprise vs. consumer): Gemini Notebook Enterprise's documented differences are governance controls, VPC Service Controls, customer-managed encryption keys, and a no-training guarantee on your data, not a different underlying model. If you're choosing Enterprise specifically hoping for a more capable model, that's not what the tier is selling; our Gemini Notebook Enterprise guide covers what actually changes.

No model picker, and no in-app indicator

No model picker has ever shipped to users, and press coverage of past upgrades has specifically noted the absence of any in-app signal that a new model is now live. After the December 2025 Gemini 3 upgrade, 9to5Google reported there was no in-app indicator that a new model was in use at all; users found out from NotebookLM's own social post, not from anything inside the product. The June 2026 upgrade didn't change that pattern. This isn't quite a story of Google never considering the idea: around the same time as the Gemini 3 rollout, in early December 2025, a Fast/Thinking-style model selector was spotted in testing by outlets tracking Google's app code. It never shipped to the public, and Gemini Notebook remains fully automatic today. If you want to know what model you're currently talking to, checking recent press coverage and this page is currently more reliable than checking inside the product itself.

What Google has published about capability, if not model name

Alongside the June 2026 upgrade, Google published self-reported benchmark figures: a 65% average win rate versus the prior chat system in internal evaluations, 78.2% on web-research-style tasks, and 69.9% on document-analysis tasks. These are Google's own numbers, not independently reproduced by a third party, so treat them the way you'd treat any vendor's self-reported benchmark: a directional claim of real improvement, not a settled, externally-verified score. No source states a context-window figure specific to Gemini Notebook's chat feature that differs from Gemini's general context-window ceiling; the context-window figures you'll see quoted for the product describe the overall model capability, not a NotebookLM-specific limit layered on top.

Model transparency: how Gemini Notebook compares to ChatGPT and Claude

This invisible-model approach is specifically a Gemini Notebook trait, not an industry norm. ChatGPT shows a model picker directly in its interface, so a user can see and choose whether they're on GPT-5.4, GPT-5.4 mini, or another available option before sending a message. Claude's interface likewise names its active model (Sonnet 5 on free and Pro, Opus 5 or Fable/Mythos 5 on higher tiers) in the model selector, and Anthropic's own release notes track version changes explicitly enough that our Claude free-tier guide can state the exact default-model switch date. Gemini Notebook is the outlier among the three in offering neither a picker nor a visible label, which is consistent with its broader design philosophy of hiding infrastructure details behind capability descriptions, but it does mean a user who cares about exactly which model is answering has to rely on outside reporting like this page rather than anything the product itself surfaces.

What about Gemini 3.7 Flash?

If you've read this site's coverage of Gemini 3.7 Flash, a coding-and-agents-focused model Google released on 13 August 2026, you might reasonably wonder why it isn't the answer to this article's question. It isn't, because Gemini Notebook and the standalone Gemini API/app release different things on different tracks: Gemini 3.7 Flash was released and priced as a developer-facing API model aimed at coding and agentic workloads, not as a consumer-chat replacement inside Gemini Notebook. No source, official or press, connects Gemini 3.7 Flash to a Gemini Notebook upgrade. That doesn't rule out a future switch, Google could adopt it later, but as of this writing Gemini Notebook's documented chat model is still Gemini 3.5 from the June 2026 upgrade, not the newer 3.7 Flash release.

Where competing coverage gets this wrong

The most common error, still visible on generic "how to use NotebookLM" tutorial pages and library research guides, is stating flatly that NotebookLM "runs on Gemini 1.5 Pro," a fact that was accurate roughly two model generations ago and has been superseded at least three times since (2.0 Flash, 2.5 Flash, Gemini 3, then Gemini 3.5). Even reference pages that get the recent history mostly right, including Wikipedia's own Gemini Notebook article, tend to skip the intermediate 1.5 Pro, 2.0 Flash, and 2.5 Flash steps entirely and jump straight from launch to Gemini 3, which flattens a version history that actually had several distinct upgrades in between.

People also ask

Does NotebookLM use Gemini Pro or Gemini Flash?

Historically, Flash-tier variants (2.0 Flash, then 2.5 Flash). Since the December 2025 upgrade to Gemini 3 and the June 2026 upgrade to Gemini 3.5, Google has not specified a Pro-vs-Flash variant for either release in its own communications.

Can I choose which Gemini model NotebookLM uses?

No. Model selection is fully automatic, and Gemini Notebook has never shown an in-app indicator confirming which model version is currently answering, even immediately after a documented upgrade.

Does the free version of Gemini Notebook use the same model as Pro?

Unconfirmed. The free tier was explicitly excluded from the June 2026 Gemini 3.5 rollout at launch, and no primary source has since stated that it was brought up to match. Don't assume parity until Google confirms it directly.

Is NotebookLM Enterprise powered by a more advanced model?

No evidence supports that. Documented differences for Enterprise are governance features (VPC Service Controls, customer-managed encryption keys, a no-training guarantee), not a different underlying model.

What model generates NotebookLM's Video Overviews?

The visual styles and animated illustrations in Video Overviews come from "Nano Banana," Google's dedicated image-generation model, a separate model from whatever text model is currently answering chat questions. Audio Overviews are a separate, older feature; no source documents them running on Nano Banana, so don't assume the two share a model.

How can I tell when Gemini Notebook has switched to a new model?

You currently can't tell from inside the product itself. Every past upgrade has been confirmed through NotebookLM's own social posts or press coverage rather than an in-app notification, so watching official announcements and independent reporting is the only reliable way to know a switch happened.