Which Model Does Gemini Notebook (formerly NotebookLM) Use? (2026)
Which model does NotebookLM use? Gemini Notebook runs on Gemini, with Gemini 3.5 named for Ultra. See what each plan gets and whether it uses Gemini 4 Argon.
Which model does Gemini Notebook (formerly NotebookLM) use? Google’s own Gemini models on every plan, but Google names a version for only one tier. Google AI Ultra and Workspace business accounts with AI Expanded Access run on Gemini 3.5 and Antigravity. Free, Plus and Pro accounts get “access to Gemini models” at rising levels, with no version number. And Gemini 4 Argon, the model in the Gemini app for Pro and Ultra, isn’t one Google has tied to Gemini Notebook. Below: what Google states for each plan, the model history in brief, and why the version matters less than your sources.
Key Takeaways
- Google names Gemini 3.5 and Antigravity for Ultra and for Workspace AI Expanded Access (Google, 2026).
- Free, Plus and Pro get “access to Gemini models” at rising levels; no version is named.
- Gemini 4 Argon is in the Gemini app for Pro and Ultra. Google hasn’t said Gemini Notebook runs on it.
- You can’t pick a model, and grounding in your sources moves answers more than the version does.
Which model does Gemini Notebook use on each plan?
Gemini on all of them, with a version named only at the top. Google said it was “upgrading NotebookLM to run on Gemini 3.5 and Antigravity”, starting with Google AI Ultra and Workspace business customers with AI Expanded Access (Google, 2026). Antigravity is Google’s coding tool, as 9to5Google describes it, and the same upgrade gives each notebook a secure cloud computer that writes and runs code on your sources. Pro has that upgraded notebook too: Gemini Notebook’s official X account says “the more powerful Notebook experience” is on all Pro accounts, but the post names no model (Gemini Notebook on X, 2026). Everything else is described as a level of access:
| Plan | Google’s plan table says | Newest model Google has named for it |
|---|---|---|
| Free (Standard) | “Access to Gemini models” | Gemini 3, announced without naming plans. |
| Google AI Plus | ”Access to Gemini models” | Gemini 3, announced without naming plans. |
| Google AI Pro | ”Higher access to Gemini models” | Gemini 3.1 Pro. The cloud-computer upgrade is on every Pro account; its model isn’t named. |
| Google AI Ultra | ”Highest access to Gemini models” | Gemini 3.5 and Antigravity. |
| Workspace business, AI Expanded Access | Set by the Workspace license | Gemini 3.5 and Antigravity. |
Plan wording from Gemini Notebook Help, checked October 26, 2026.
Google One’s plan page puts it the same way. It promises “More access”, “Expanded access” and “Higher access to features and models in Gemini Notebook” for Plus, Pro and Ultra in turn, and never names the model (Google One, 2026).
Is Gemini Notebook on Gemini 4 Argon?
Not that Google has said. Gemini 4 Argon is Google’s frontier model, and Google made it available in the Gemini app for Google AI Pro and Ultra subscribers (Google, 2026). Google One lists Argon for the Gemini app, AI Mode in Search, Google Antigravity and Google AI Studio, yet the Gemini Notebook line on the same plan cards names no model (Google One, 2026). Neither Google’s Argon posts nor Gemini Notebook’s help pages mention Argon. So if a video or post says your notebooks run on Argon, treat it as a guess. The Gemini app’s own model choice differs by plan, from mostly Flash-Lite on the free tier to Argon on Pro and Ultra, and the plan comparison linked below lists it.
Does Gemini Notebook use Gemini?
Yes, and only Google’s own models. The LM in NotebookLM stood for language model: Google’s 2023 launch post said the team built it “knowing that you would have a powerful language model at its core: hence the LM” (Google, 2023). Every model Google has named for it since December 2023 is a Gemini model. If you’re new to the tool itself, our explainer on what Gemini Notebook is and how it works covers the basics.
How has the Gemini Notebook model changed since 2023?
Eight named steps, from an unnamed language model to Gemini 3.5. Here’s the record, one row per model, oldest first:
| Month | Model | Source |
|---|---|---|
| July 2023 | An unnamed “language model” | |
| December 2023 | Gemini Pro | |
| June 2024 | Gemini 1.5 Pro | |
| December 2024 | Gemini 2.0 Flash (experimental) | |
| May 2025 | Gemini 2.5 Flash, for Q&A | NotebookLM on X, via BleepingComputer |
| December 2025 | Gemini 3, variant not named | NotebookLM on X, via 9to5Google |
| February 2026 | Gemini 3.1 Pro, on Pro and Ultra | |
| June 2026 | Gemini 3.5 and Antigravity, on Ultra and AI Expanded Access |
The Gemini 3 row has a gap. 9to5Google wrote that the announcement “does not specify the specific model, but NotebookLM has historically used the Flash variants”, which is an educated guess rather than a statement from Google. And NotebookLM didn’t launch on Gemini 1.5 Pro, a common mix-up: that model came almost a year after launch.
Read down the table and one pattern stands out. Up to Gemini 3, Google’s announcements didn’t split models by plan. From Gemini 3.1 Pro on, every newly named version has gone to paying tiers first, and only the top tier has a version on record today. For everything else that changed month by month, see our Gemini Notebook changelog.
Do free and paid plans get different models?
Yes, at least for the newest ones. When Gemini 3.1 Pro launched, Google said it was “now available on NotebookLM exclusively for Pro and Ultra users” (Google, 2026). Gemini 3.5 is named for Ultra. Google’s plan table puts the rest in terms of access: the free tier and Plus get “Access to Gemini models”, Pro “Higher access” and Ultra “Highest access” (Gemini Notebook Help, 2026). What the table doesn’t say is which model a free account runs, or whether “higher access” means a newer model, more of the same one, or both. Gemini Notebook’s plan card on Google’s own site is just as vague: Ultra gets “Highest limits and best model capabilities”, and no plan gets a model name (Google, 2026).
Are paid answers faster or better?
Google hasn’t published speed or answer-quality figures by plan. What it has published is how much you can use. Usage limits factor in “the models and features you use”, and Plus gets two times the standard limits, Pro four times, and Ultra 5x or 20x Pro’s (Gemini Notebook Help, 2026). Our usage limits guide explains how that works. Google’s only quality numbers compare the Gemini 3.5 system with the one before it, not plan with plan. The upgrade won “an average win rate of over 65%” in side-by-side tests, 69.9% on large document analysis and 78.2% on web research and source discovery (Google, 2026). Those are Google’s own evaluations.
Capacity is the difference you can see in writing. A free notebook holds 50 sources, Plus 100, Pro 300 and Ultra up to 600 (Gemini Notebook Help, 2026). For prices and everything else each plan adds, compare Plus vs Pro vs Ultra; for what the free tier covers, see whether Gemini Notebook is free.
Can you choose a model in Gemini Notebook?
No. Gemini Notebook has no model menu. Configure Chat is the only chat setting: it picks a conversational style (Default, Learning Guide or Custom) and a response length (Gemini Notebook Help, 2026), and it switches connected apps on or off for that notebook (Gemini Notebook Help, 2026). The Gemini app is different. Its model menu offers Flash-Lite, Flash and Pro depending on your plan, each at a low, medium or high effort level (Gemini Apps Help, 2026).
Upgrades also arrive quietly. When Gemini 3 landed, 9to5Google noted that “there’s no in-app indicator that signals a new model is in use” (9to5Google, 2025). So a prompt that gave you a sharp answer in May can answer differently in July, with no version change on screen. The practical fix is to keep the prompts that work and save the answers you rely on.
Keep what worked before the next model swap
Model upgrades land without notice. Kortex keeps your best prompts one click away and exports chats, notes and sources to PDF or Markdown, so a good answer outlives the model that wrote it. Free on Chrome.
Which models make Audio Overviews, Video Overviews and slides?
Different ones, and Google names them more readily than the chat model. The Studio outputs mix Gemini with Google’s image, video and speech models:
| Output | Model Google names | Source |
|---|---|---|
| Chat and Q&A | Gemini, 3.5 on top plans | Google, 2026 |
| Audio Overviews | Google DeepMind speech generation research; Gemini 3.8 Flash TTS is in Gemini Notebook, output not specified | Google DeepMind, 2024; Google, 2026 |
| Video Overviews (Explainer, Brief) | Nano Banana, Gemini’s image model, for the illustrations | Google, 2025 |
| Cinematic Video Overviews | Gemini 3, Nano Banana Pro and Veo 3 | Google, 2026 |
| Infographics and Slide Decks | Nano Banana Pro (Gemini 3 Pro Image) | Google Workspace Updates, 2025 |
Audio Overviews are the odd one out. Google DeepMind says it “helped develop” them and describes the “speech generation research underpinning” them, without naming a Gemini version (Google DeepMind, 2024). Google’s newest speech model, Gemini 3.8 Flash TTS, is the closest thing to a name. Google calls it one of “our most expressive audio generation models yet” and lists it “For everyone: In Gemini Notebook”, without saying which outputs use it (Google, 2026). Chat and audio have run on separate models before, too: when chat moved to Gemini 2.5 Flash, the NotebookLM team said the change would “currently only impact the Q&A” (BleepingComputer, 2025). Video is better documented. Explainer and Brief videos use Nano Banana for illustrations (Google, 2025). Cinematic videos, meanwhile, combine Gemini 3, Nano Banana Pro and Veo 3, with Gemini acting as “a creative director” (Google, 2026). Infographics and slide decks run on Nano Banana Pro, which Google also calls Gemini 3 Pro Image (Google Workspace Updates, 2025). Our Video Overviews guide compares every format.
Does the model version change your answers?
Less than grounding does. Source grounding means the answer has to come from documents you supplied, not from the model’s training data. Google’s help center says “chat responses in Gemini Notebook only use data from your sources” (Gemini Notebook Help, 2026). That setup does more for accuracy than a version bump. A 2025 study gave three tools a reporting task over 300 documents: 13% of NotebookLM outputs contained a hallucination, against about 40% for Gemini and ChatGPT (Hagar et al., arXiv, 2025). NotebookLM and the Gemini chatbot both ran on Google’s Gemini models, yet the grounded tool made about a third as many errors. Across all three tools, most errors weren’t invented facts, either. The authors found “interpretive overconfidence” instead, such as turning an attributed opinion into a general statement. A newer model won’t fix that for you. Checking the citations will.
Why your source selection matters more than the context window
Because no model reads a full notebook at once. Gemini Notebook’s chat uses “the full 1 million token context window of Gemini” on every plan (Google, 2025). Google’s developer docs put 100 tokens at “about 60-80 English words” (Google AI for Developers, 2026). In other words, the window holds roughly 600,000 to 800,000 words.
Now do the maths on a full notebook. Each source can run to 500,000 words (Gemini Notebook Help, 2026). A full free notebook, with 50 sources, could therefore hold 25 million words, about 30 to 40 times the window. A full Pro notebook, at 300 sources, could hold 150 million, about 190 to 250 times. So Gemini Notebook picks passages first. Google says that when a notebook has many sources, it “retrieves the most relevant information based on your question” (Gemini Notebook Help, 2026). That makes the source checkboxes your real dial. For example, ask about one contract with only that contract ticked, and the retrieval step has less to sift. Then click a citation or two before you reuse the answer.
How do you pair your preferred AI model with Gemini Notebook?
Let your own assistant ask the notebook. Gemini Notebook has no public consumer API, so Claude, ChatGPT or Codex can’t reach your notebooks on their own. Kortex MCP is the bridge we built for that, so weigh this section as the maker’s view. Your assistant sends the question through Kortex, Gemini Notebook answers from your sources, and the answer comes back with its citations (Kortex MCP docs, 2026). The model you prefer then does the reasoning, writing or coding on top, while Gemini Notebook keeps doing the grounding.
People clearly have favourites. In the Kortex MCP waitlist survey, then in private beta (776 respondents, Sep 19 to Oct 1, 2026), 126 people named Claude in their answers, 69 ChatGPT and 61 Gemini. Separately, 110 of the 776 asked to query their notebooks for cited answers from their assistant.
Here’s what the split looks like in a single message to Claude:
Ask my "Vendor contracts" notebook what each contract says about termination notice, with citations. Then draft a one-page summary table and flag any contract that differs from our standard 30 days.Gemini Notebook answers the first half from the contracts, citations attached. Claude then builds the table and flags the outliers. Need several notebooks? The ask_library tool asks up to three at once and keeps each answer separate. Kortex MCP is included with Kortex Pro ($9.99 a month, or $4.92 a month billed yearly) and Lifetime ($200 once). Our guide to connecting Gemini Notebook to Claude and ChatGPT walks through the setup.
Your model for thinking, your notebooks for facts
Ask any notebook from Claude, ChatGPT or Codex, get Gemini Notebook's cited answer back, and keep working in the same chat. Kortex MCP is included with Kortex Pro and Lifetime.
How can you tell which model your notebook uses?
From your plan, since the screen won’t say. Gemini Notebook shows no model name anywhere, but it does show your tier: upgraded accounts get a distinct badge next to the profile picture (Gemini Notebook Help, 2026). Match the badge to the plan table above. Ultra means Gemini 3.5 and Antigravity. Pro means the upgraded notebook with its cloud computer, on a model Google doesn’t name. Free or Plus means “access to Gemini models”. The short version: Gemini Notebook runs on Gemini, the top plan has the newest version Google names, and your sources decide more about each answer than the model does.
Frequently asked questions
Which model does NotebookLM use?
Google’s Gemini models. Google names Gemini 3.5 and Antigravity for Gemini Notebook (formerly NotebookLM) on Google AI Ultra and on Workspace business accounts with AI Expanded Access. Free, Plus and Pro get “access to Gemini models” at rising levels, with no version named, and Google hasn’t said which Gemini 3.5 variant it uses.
Does NotebookLM use Gemini?
Yes. Every model Google has named for it since December 2023 is a Gemini model: Gemini Pro, then Gemini 1.5 Pro, 2.0 Flash, 2.5 Flash, Gemini 3, 3.1 Pro and Gemini 3.5. The LM in its old name stood for language model.
Does Gemini Notebook use Gemini 4 Argon?
Google hasn’t said so. Gemini 4 Argon is in the Gemini app for Google AI Pro and Ultra, and Google’s plan pages also list it for AI Mode in Search, Antigravity and AI Studio. Their Gemini Notebook rows, however, name no model, and Gemini Notebook’s help pages don’t mention Argon.
Which Gemini version does the free plan use?
Google doesn’t say. Its plan table lists “Access to Gemini models” for free and Plus, “Higher access” for Pro and “Highest access” for Ultra. The last model Google named without plan limits was Gemini 3. After that, Gemini 3.1 Pro went to Pro and Ultra, and Gemini 3.5 to Ultra.
Can you choose the model in Gemini Notebook?
No. Configure Chat sets a conversational style (Default, Learning Guide or Custom), a response length and which connected apps a notebook may use, but there’s no model menu. Upgrades also arrive without an in-app notice, so you can’t see which version answered.
Which model makes Audio Overviews?
Google doesn’t name one. Google DeepMind says it helped develop Audio Overviews and that its speech generation research underpins them. Google also lists Gemini 3.8 Flash TTS, which it calls its most expressive audio generation model, as available to everyone in Gemini Notebook, without saying which outputs use it.
Can I use Claude or ChatGPT with my notebooks instead?
Not inside Gemini Notebook, but you can pair them. With Kortex MCP, Claude, ChatGPT or Codex sends your question to a notebook and gets Gemini Notebook’s cited answer back, then does the reasoning or writing itself. Kortex MCP is included with Kortex Pro and Lifetime.
Editorial note: fact-checked against the official help pages and docs linked above on October 26, 2026. Spotted something out of date? Contact us and we will fix it, or read more about Kortex.
