Product3 distinct publishers3 min readPublished
From September 2, the free tier's three overviews a day becomes a share of compute that varies with prompt complexity and source count, and every paid tier is quoted as a multiple of a base Google has not published.
The Product Desk · Product desk
Compiled by The Product DeskSomething wrong?How this is made
The only price tag in this system is a bar that fills up underneath a Studio generation, and the more of it that is filled, the higher the expected cost of the output [10]. Phandroid reports the interface tracks usage in real time [16]. What it will not tell you is how many Video Overviews you have left this week, because after September 2 that number stops existing.
The multipliers work out like this: AI Pro is quoted at 4x standard limits and the top Ultra tier at 20x higher than Pro, which works out to 80x standard [5][1]. Eighty times what is the open question, because none of the reports puts an absolute figure on standard limits [4]. The scheme being retired did carry absolute figures. Android Police lists three audio or video overviews and 10 reports or quizzes per day on free, against 5,000 chat queries and 200 overviews a day on Ultra [7], a spread of roughly 67x on overviews [2]. You could put that in a runbook. You could tell a client how many decks they were getting.
The five-hour cycle is the other part that reads better in a blog post than on a calendar. Google frames it as continuous access, saying that because limits refresh every five hours rather than daily you can keep working through the day [15]. Five does not divide 24, so the windows drift: 4.8 of them per day, and the 3:00 PM reset in Google's own interface copy [9] becomes 8:00 PM on the next cycle and 1:00 AM after that [3]. That drift makes the reset time useless for scheduling.
Teams assume a cost meter teaches users to learn what the expensive features cost and spend deliberately. In practice, users take whatever the product offers instead, which Google has already pre-built, since the notebook will suggest alternative outputs when a first choice goes over the limit [11]. Users take the substitute without weighing it, and it happens without anyone filing a ticket.
The mechanism worth arguing about is source count. The number of sources attached is an input to the cost calculation [4], which means the notebook you have been feeding for three months costs more per question than the one you opened this morning. Research accumulates. Under a per-day feature count, that accumulation was free: the tenth overview cost the same as the first [1].
So, two axes. Down one side, can this output wait five hours. Across the top, is this notebook heavy or light in attached sources. Light and can-wait was never the problem, and heavy and can-wait is where the deferral queue earns its keep. Heavy and deadline-bound is the quadrant that breaks, and the available defences are producing it early or buying a larger multiple of a base you cannot inspect. The measurement worth having is not the number on the pricing page but the count of finished outputs your team gets through in a normal week, taken once before September 2 and again a fortnight after. That count is the ceiling you will be asked about on a Friday.
Ranked by verification strength, evidence, and original report placement.
Gemini Notebook is switching to compute-specific usage limits, replacing the set number of feature generations per day that depended on the user's Google AI tier.
The compute-based usage limits begin rolling out on September 2, 2026, to consumer accounts on web and mobile.
Google says the overall usage limit will factor in prompt complexity, chat length, number of sources, and the features used.
Google's support page says the quota refreshes every five hours until the user reaches a weekly limit, and that users can upgrade for higher AI limits.
Gemini Notebook will suggest alternative outputs if a user's first choice surpasses their limit.
Distinct publishers with included, body-backed reporting in this cluster.
2 articles · August 28, 2026
1 article · August 28, 2026
1 article · August 29, 2026
Follow any of these and your For You feed starts watching them — no settings page required.
product
Google gives US students a $200 AI plan free, plus a renewal date and 5TB of switching cost3 distinct publishers
invest
Scalable Capital puts ChatGPT, Claude and Grok inside the European order ticket2 distinct publishers
product
OpenAI shipped a teen ChatGPT. The over-65 cohort doubled to 23% and got nothing.1 distinct publisher
security
The nationalization argument is really a vendor-continuity memo1 distinct publisher
Evidence-backed comparisons of source perspectives and observed adoption signals. Read the methodology
Which Builder, Operator, and Investor concerns the observed source mix emphasized—not a truth score.
Evidence, demonstrated adoption, hype gap, incentives, and confidence are assessed independently, each on its own current evidence. How these are measured.
One origin, quoted accurately
Three publishers, four items, and a single source of truth: Google's announcement plus a support page that Android Police quotes verbatim. That is why the date, the five-hour cadence and the tier ladder match everywhere — agreement here is transcription, not corroboration. Two of our four items are the same 9to5Google post. The documentary record is clean on mechanics and empty on magnitude: nothing states what 'standard limits' actually is.
Dated, not yet shipped
Nothing has switched over. What lifts this above zero is precedent rather than usage: Android Police documents the same compute meter and five-hour window running on the Gemini app since May, and Google has committed to a specific date across all consumer web and mobile accounts. Until September 2 this is a calendar entry, and no report contains a single user's experience of the new limits.
Flexibility priced in an unknown unit
Phandroid carries Google's promise that the change 'gives you more choice' and lets you 'keep working throughout the day', and no one can check it: 4x standard is only meaningful if you know standard. The one hard before-and-after in our coverage comes from Android Police — three overviews a day free, 200 on Ultra — and it is a ceiling being removed, not a floor being guaranteed. Android Police's own 'more complicated' framing is the sole counterweight, which is why the gap is a lean rather than a chasm.
Upsell wired into the meter
The support sentence Android Police quotes ends exactly where these sentences usually end: come back later, or upgrade for higher AI limits. Charging by compute rather than by generation moves the cost of an expensive Video Overview onto the user's own budget, and defining every tier as a multiple of an unpublished base leaves the denominator in Google's hands. On the reporting side, these are three Google-beat outlets working from one announcement, and the one closest to the company's language prints no numbers at all.
Consistent, untested, pre-launch
I would bet on the mechanics and not on the outcome. The three accounts never contradict each other, and where they differ it is a dropped item — Android Police's support-page list names models where 9to5Google's names attached sources. But the interface strings and the Studio cost bar rest on one post that appears twice in our coverage, the weekly ceiling on one publisher, the plan prices on the same one, and the whole thing is described before anybody has hit a limit under it.