AI limit UI: preview cost, 5-hour resets, generate later

A user asks for a slide deck, hits an AI limit, and sees no reason for the stop. A better limit UI treats output as a compute budget, with three moves:

• Preview the cost before heavy generation.
• Use 5-hour resets with a weekly ceiling.
• Let heavy outputs wait with a ready alert.

That keeps the user working instead of forcing a choice between waiting and leaving.

Make the cost visible before the click

Put a thin cost-preview bar or label next to the generate button, and let it update as the prompt changes. It should estimate how much of the current window the output will consume, so the user can compare a short summary, a full draft, and a slide deck before choosing. If a full deck will consume most of the window, the user can ask for a shorter answer first. The request becomes a budgeting decision, and the product avoids a surprise stop.

Do not lead with raw tokens. Users do not own the model's metering, so raw tokens are hard to judge. Show relative budget: “small,” “medium,” “heavy,” or a percentage of the window. If the request would exceed the current allowance, say so before the click, not after the failure.

API pricing makes the budget concrete. Anthropic's Claude Sonnet 5 launch price was $2 per million input tokens and $10 per million output tokens through August 31, then $3 per million input tokens and $15 per million output tokens. Google's API pricing for Gemini 3.6 Flash was $1.50 per million input tokens and $7.50 per million output tokens, while Gemini 3.5 Flash-Lite was $0.30 per million input tokens and $2.50 per million output tokens. The designer's job is to translate that spread into choices the user can make before spending it.

The strongest public example is Gemini Notebook, which will switch to compute-based usage limits on September 2, 2026. Gemini Notebook calculates usage from prompt complexity, the models and features used, chat length, and specific features. The cost preview should use the same inputs, because a long chat with a heavy feature is not the same request as a short question.

Use 5-hour resets with a weekly ceiling

Gemini Notebook's new limits reset every five hours rather than once each day. The quota renews every 5 hours until the user reaches a weekly ceiling. The weekly ceiling is the long-range guardrail for heavy work. A five-hour window keeps work moving through a workday without making the user wait until midnight. The timer should be a status line near the generate control, naming the next refill and the weekly ceiling, not a banner that interrupts the work.

The mistake is to make the timer a wall. If the UI stops at a limit message and hides the next refill, the user leaves. The redesign is to show what renews, when it renews, and what remains after the cap is reached. A daily cap can feel arbitrary when work arrives in short blocks; the return should be obvious, not accidental.

Let heavy outputs wait with a ready alert

Offer deferral when the request is too heavy for the current window. Users who hit the limit can postpone creating Video Overviews or Slide Decks until later. The result should be a generate-later button, not an upgrade prompt.

A postponed generation can require a couple of hours, and an alert can be sent when it is ready if the user has notifications on. The deferral path should make that wait explicit: queue the job, show an estimated readiness time, and send a ready alert only if the user opted in.

Gemini Notebook will monitor usage and offer different output options when the selected output would go over the limit. The fallback should give the user a choice: shorten the scope, downgrade the output, or defer the job. The queue should keep the job visible until it is ready.