update · TowCue Editorial Team

Gemini Notebook flexible usage limits: what changes on September 2

Gemini Notebook is moving from daily feature caps to compute-based limits with five-hour refreshes. See what changes, who benefits, and what remains unconfirmed.

Original editorial contentSources verifiedLast reviewed: 2026-08-28

Quick answer

Google announced a fundamental shift in how Gemini Notebook manages usage limits, effective September 2, 2026. The product moves from daily feature caps (e.g., "3 Video Overviews per day") to compute-specific limits that refresh every 5 hours until a weekly cap is reached.

The new system weighs prompt complexity, chat length, source count, and features used (Video Overviews, Slide Decks, Deep Research, code execution, etc.) against a compute budget. Heavy tasks that exceed the current 5-hour window can be deferred with "Generate later" — they complete automatically in the background and notify the user when ready.

Rollout begins September 2, 2026 for consumer accounts on web and mobile. Enterprise/Workspace accounts follow their admin-controlled timeline.

For broader product context, read TowCue's complete Gemini review, compare this change with the Gemini connected-apps update, or use the AI tool selection guide before changing your workflow.

What changed

AspectOld model (until Sep 1)New model (from Sep 2)
Limit typeDaily feature caps (per artifact type)Compute-specific budget (unified)
Refresh cadenceDaily (24-hour reset)Every 5 hours (rolling) + weekly cap
Cost factorsOne artifact = one countPrompt complexity, chat length, source count, features used
Heavy task handlingBlocked until next daily reset"Generate later" — queues task, auto-completes, notifies
VisibilityPost-generation "limit reached"Real-time usage bar + "Limit resets at 3:00 PM" warnings
Tier multipliersFixed per-artifact countsStandard / 2× (Plus) / 4× (Pro) / 5–20× (Ultra)

Official source: Google Blog — "We're introducing flexible usage limits for Gemini Notebook" (August 28, 2026, by Yesul Shin, Product Manager, Gemini Notebook)

Support doc: Manage your Gemini Notebook usage limits (effective September 2, 2026)

Why this matters: a shift in usage logic

The old model treated every artifact equally — one Video Overview = one count, regardless of whether it was a 30-second explainer or a 20-minute cinematic deep dive. The new model acknowledges that compute cost varies dramatically:

  • Simple chat query with 2 sources → low compute
  • 50-source notebook, long chat, Deep Research + Video Overview + Slide Deck → high compute
  • Code execution in secure cloud computer → variable compute

By pricing in compute, Google aligns limits with actual resource consumption, not arbitrary artifact counts. This also means light users get more mileage (many simple queries per window), while heavy workflows consume budget proportionally.

The "Generate later" queue: asynchronous workflows arrive

This is the most workflow-relevant change. When a Studio generation (Video Overview, Slide Deck, Audio Overview, Report, etc.) would exceed your current 5-hour compute budget:

  1. You see the expected cost via a usage bar before confirming.
  2. If over budget, you can choose "Generate later" instead of canceling.
  3. The task queues in the background.
  4. At the next 5-hour refresh (or when budget allows), it auto-completes.
  5. You get a notification (if enabled) when the artifact is ready.

This turns Gemini Notebook from a synchronous tool (you wait, you watch, you hit a wall) into an asynchronous pipeline — closer to how CI/CD or batch rendering works. For researchers and analysts running multiple heavy artifacts per session, this eliminates the "stop and wait" friction.

Limitation (per support doc): "Generate later" is web-only as of the September 2 launch. Mobile users must wait for the refresh or upgrade.

Tier structure: compute budget scales with plan

PlanCompute budget (relative to standard)
No plan (free)1× (standard)
AI Plus ($4.99/mo)
AI Pro ($19.99/mo)
AI Ultra ($99.99 / $199.99/mo)5× or 20× Pro (depending on Ultra tier)

The multipliers apply to the compute budget, not artifact counts. A Pro user gets ~4× the compute budget of a free user per 5-hour window, and hits the weekly cap later.

Weekly cap: The 5-hour refreshes accumulate toward a weekly ceiling. Once the weekly cap is reached, no further refreshes occur until the week rolls over. Google has not published exact weekly numbers; they are "subject to change."

Visibility: real-time feedback replaces surprise blocks

  • Chat footer: "You're almost at your AI usage limit. Limit resets at 3:00 PM."
  • Studio panel: Expected cost bar fills as you configure a generation (e.g., adding sources, choosing cinematic video).
  • Settings → Usage: Detailed breakdown of current window usage, weekly progress, next refresh time.

This transparency lets you plan heavy work (e.g., "I have 70% budget left, I'll queue the Video Overview for later and finish the Report now").

What stays the same (capacity limits)

The following do not reset with the 5-hour/weekly cycle — they are structural caps:

LimitFreePlusProUltra 20 TBUltra 30 TB
Notebooks per user100200500500500
Sources per notebook50100300500600
Words per source500,000500,000500,000500,000500,000
Local upload size200 MB200 MB200 MB200 MB200 MB

These capacity limits are unchanged by the September 2 update.

What we don't know (unconfirmed)

  • Exact weekly compute cap per tier (Google: "subject to change")
  • Per-prompt compute ceiling (I/O 2026 introduced a cap after backlash; not confirmed if carried over)
  • Enterprise/Workspace rollout timeline (admin-controlled, likely after consumer)
  • Whether "Generate later" will come to mobile (currently web-only)
  • Notification granularity (push, email, in-app only?)
  • Interaction with code execution (secure cloud computer compute cost not documented)

TowCue take

This update signals a broader industry shift: AI products are moving from "feature gates" to "compute meters."

For individual researchers, the 5-hour refresh + "Generate later" queue is a genuine workflow improvement — you can keep a notebook open all day, queue heavy artifacts, and collect them at natural breakpoints (lunch, end of day).

For teams and power users, the compute budget model demands new habits:

  • Audit your typical session: how many sources, how long the chat, which Studio outputs you chain.
  • Prefer "Generate later" for Video Overviews/Slide Decks — they're the heaviest per unit.
  • Use the usage bar as a design tool: configure the output, check the bar, decide now vs. later.
  • Watch the weekly cap on multi-day projects; a 5-hour refresh doesn't help if the week is exhausted.

For Google, this is also a pricing lever: compute budgets are easier to monetize (top-up credits, per-token overages) than artifact counts. Expect "pay-as-you-go compute credits" to appear for Pro/Ultra tiers, as hinted at I/O 2026.

Verify your current tier's effective budget on September 2 via Settings → Usage. The old daily artifact counts will no longer apply.

Research sources

Research sources

Turn this intelligence into a reusable Cue

Related decision guides