Yes, Gemini has limits on every tier including the paid ones — but the way they work changed fundamentally in May 2026, and most of the numbers still circulating online describe a system Google no longer uses.
The short answer
Gemini no longer counts prompts. It measures compute.
There is no message counter to watch and no fixed daily allowance to ration. Instead, each request draws from an allowance based on how demanding it actually is, and that allowance refreshes every five hours inside a weekly ceiling.
What replaced the daily prompt counts
Until May 2026 the limits were simple and widely quoted:
| Tier | Old daily limit |
|---|
| Free | 5 prompts/day |
| Pro | 100 prompts/day |
| Ultra | 500 prompts/day |
Those numbers are obsolete. If you find them on a page today, that page has not been updated.
Google now publishes multipliers rather than counts:
| Tier | Allowance |
|---|
| Free | Standard |
| AI Plus | 2× standard |
| AI Pro | 4× standard |
| AI Ultra | 5×–20× Pro, depending on tier |
Google has not published the absolute value of "standard", which is the honest frustration with the new system — you can compare tiers to each other but not predict where you will run out.
What actually consumes your allowance
Four things, and this is the part worth internalising because it is where the behaviour change is:
- Prompt complexity. A hard reasoning question costs more than a simple lookup
- Which model you use. The heavier models draw down faster
- Features invoked. Deep research, image generation, code execution and file analysis each cost more than plain chat
- Conversation length. A long thread re-reads its own history on every turn, so the same question costs more at message 60 than at message 3
That last one catches people out most. If you are hitting limits sooner than expected, long-running conversations are usually why.
Why Google changed it
A fixed prompt count charges the same for "what is the capital of France" as for a forty-file codebase analysis, which is wrong in both directions — it wastes allowance on trivial requests and undercharges expensive ones. Compute-based limits are more honest per request.
The cost is predictability. Under the old system you knew you had five prompts. Now you know you have an allowance that refreshes in five hours, and you find its edge by hitting it.
Google also tightened free access to Gemini 3 Pro separately, which is generally read as a demand-management move rather than a policy shift.
How to make your allowance last
- Start a new chat for a new topic. The single most effective habit — it stops you paying to re-read irrelevant history
- Match the model to the task. Do not spend a heavy reasoning model on a factual lookup
- Batch related questions into one well-formed prompt rather than ten follow-ups
- Save deep research and file analysis for when you actually need them
- Remember the five-hour refresh. Hitting a wall is temporary, not a day lost
So which tier do you need?
Free is genuinely usable for occasional questions, and it is where the five-hour refresh matters most — you will meet the ceiling, but not for long.
AI Pro at 4× is the sensible choice for daily working use and is where most people who use Gemini for real work land.
AI Ultra only makes sense for sustained heavy use — long research sessions, large file analysis, or all-day agentic work. The 5×–20× range over Pro is wide, and worth checking against current documentation before paying for it.
The honest summary: the limits are real on every tier, they are no longer countable in advance, and the most reliable way to stretch them is shorter conversations rather than fewer questions.
