Gemini Notebook Is Switching to Compute-Based Usage Limits

Google is changing how Gemini Notebook measures access to AI features on September 2, 2026. Instead of assigning separate quotas for chats, Audio Overviews, Video Overviews, and other tools, the service will use a shared compute-based usage system. 67

That means the number of actions you can complete will depend less on counting requests and more on how demanding those requests are. A short text question should generally consume less of the available budget than a long conversation built around many sources or a resource-intensive multimedia output.

What is changing on September 2?

The current system is organized around feature-specific limits. The new system combines usage into an overall allowance that reflects the computational demands of the work.

Google says the meter will consider:

• Prompt complexity
• Chat length
• The number of sources in a notebook
• The models and features used

This creates a single pool that users can spend across supported Notebook features, rather than forcing them to manage entirely separate generation counts. 67

Five-hour refreshes replace daily resets

The replenishing usage limit will refresh every five hours instead of resetting once per day. However, the refreshes continue only until the user reaches an overall weekly limit. 6

This should make the system more flexible during the day: reaching a short-term limit does not necessarily mean waiting until the next day. It also means that users who make intensive use of Notebook over several refresh cycles may eventually encounter the weekly ceiling.

Google has not published one universal number of prompts that every user can expect. Under a compute-based model, different requests consume different amounts of capacity, so a fixed prompt count would not describe every user’s allowance accurately. 6

How the limits vary by Google AI plan

Google’s published plan structure scales usage against standard access:

• Without a plan: Standard limits
• Google AI Plus: 2× standard limits
• Google AI Pro: 4× standard limits
• Google AI Ultra: Either 5× or 20× the Pro allowance, depending on the Ultra subscription

Google’s plan information also describes the Ultra tiers as offering five or 20 times the Pro plan’s usage limits. 610

These are relative allowances, not a promise that a particular plan supports a fixed number of videos, chats, or other outputs. The actual amount of work a user can complete will depend on the mixture of tasks and features they use.

Notebook will show usage and warn before a limit

The new system is designed to make consumption more visible. Gemini Notebook will show usage information and the next reset time in the chat interface. Users may see warnings when they are approaching a limit or have reached it. 67

For Studio outputs, Google says Notebook will show an expected usage-cost indicator. If a requested result would use more of the remaining budget than is available, the product can suggest an alternative that requires fewer resources. 67

That guidance matters because users will no longer be able to estimate capacity simply by counting how many feature generations remain. The interface becomes the main way to understand how much of the shared budget a task is likely to consume.

What happens to expensive Video Overviews and Slide Decks?

Users who run out of active allowance can choose to generate some Studio artifacts later, including Video Overviews and Slide Decks. The deferred-generation option is available on the web and may take hours; Notebook can notify the user when the result is ready. 67

This gives users an alternative to abandoning a resource-intensive task immediately. It does not, however, make the output free of usage cost: the request still has to be processed when capacity becomes available.

Why Google is moving away from request counts

A single request is a poor measure of resource consumption when AI products handle increasingly long contexts, agentic workflows, advanced models, research tasks, and multimedia generation. Two prompts can look similar from a user’s perspective while requiring very different amounts of computation.

Google has already introduced a similar compute-based approach in the Gemini app, where limits factor in prompt complexity, the models and features used, and chat length. Those limits also refresh every five hours until a weekly limit is reached. 1

GitHub has made a related change in Copilot: from June 1, 2026, its plans moved from premium-request counting to usage-based billing, with costs tied to the model and tokens consumed and allowances provided through GitHub AI Credits.

The direction is consistent across these products: as AI systems perform longer, more autonomous, and more expensive tasks, providers are moving from a simple “one request equals one unit” model toward metering that better reflects underlying usage.

What remains unclear for heavy Notebook users

The change should give light users more flexibility if simple interactions consume relatively little of the shared allowance. Heavy users of Audio Overviews, Video Overviews, or other demanding features may find planning more difficult, because Google has not published feature-level cost weights or a universal conversion from one feature to another.

The practical impact will therefore depend on how the meter behaves in real use. Google’s documentation notes that limits can change based on availability and testing, so the published multipliers should be treated as the current framework rather than a permanent guarantee. 6

For now, the clearest expectation is that Gemini Notebook will become more flexible but less predictable: users can spend their allowance across features and recover access every five hours, yet the amount of work remaining will depend on the complexity of what they ask Notebook to do.