How many tokens a month do Gemini plans actually give you?
Google publishes compute multipliers and 5-hour windows, not token numbers. Community estimates place Gemini 4 Neon at ~800M (Pro) to ~19B (Ultra 20x), but agent workload, cache hit rates, and weekly caps change effective yields.
Google publishes no monthly token number and no prompt count for Google AI Pro or Ultra on personal accounts. The official meter is compute: an allowance that refreshes every 5 hours inside a weekly ceiling, plus a multiplier per plan. Anyone quoting “how many tokens a month” is converting that meter. This page separates the official rules from those conversions.
Takeaway
Open Gemini Apps limits and Settings → Usage Limits on your own account. Google tells you the window, the weekly cap, and the plan multiplier. It does not tell you 800 million tokens.
Community estimates for running Gemini 4 Neon all month sit around 800 million tokens on AI Pro, ~5 billion on Ultra 5x, and ~19 billion on Ultra 20x. Those are user calculations under hypothetical continuous use, status partial. Agent workload, cache hit rates, and the weekly ceiling change what you actually get. The official personal plans list the US stickers; quota limits explained covers the 5-hour plus weekly shape.
Scope
Personal Google AI Plus / Pro / Ultra in the United States, checked 20 November 2026. Consumer Gemini apps and Google Antigravity. The Developer API is a separate invoice at list rates.
What Google actually meters
Since May 2026, Gemini Apps limits are compute-based. Google’s help page, checked 20 November 2026, says the allowance refreshes every 5 hours until you reach a weekly limit. Prompt complexity, model, features, and chat length all count. Deep Research and video cost more than a short chat. Limits may change without notice, including for capacity.
Google publishes multipliers only:
| Plan (US personal) | Official usage multiple | Context window in the app | US list price, 20 Nov 2026 |
|---|---|---|---|
| Free (no AI plan) | Standard | 32k tokens | $0 |
| Google AI Plus | 2× standard | 128k tokens | $4.99 / month |
| Google AI Pro | 4× standard | 1 million tokens | $19.99 / month |
| Google AI Ultra 5x | 5× Pro | 1 million tokens | $99.99 / month |
| Google AI Ultra 20x | 20× Pro | 1 million tokens | $199.99 / month |
Prices are from gemini.google/subscriptions with the region set to United States. Open that page in your country before you pay. Tax is extra. The same bill bundles Google One storage; Antigravity paid access is this subscription, not a second SKU.
Official source: Google Gemini Apps Help, Gemini Apps limits & feature availability, checked 20 November 2026.
Community token conversions
A conversion from those multipliers into tokens is a user estimate. Status on this page is partial because Google does not publish the token column.
On 20 November 2026, @ohayou792 posted that running Gemini 4 Neon all month comes out to about:
| Plan | Community estimate (Neon, all month) |
|---|---|
| Google AI Pro | ~800 million tokens |
| Google AI Ultra 5x | ~5 billion tokens |
| Google AI Ultra 20x | ~19 billion tokens |
Check that against the official multipliers, using 800 million on Pro as the baseline:
| Plan | Official multiple of Pro | 800M × that multiple | Community claim | Gap |
|---|---|---|---|---|
| Ultra 5x | 5× | 4 billion | ~5 billion | ~1 billion higher |
| Ultra 20x | 20× | 16 billion | ~19 billion | ~3 billion higher |
They sit in the same ballpark. They are not exact. Continuous-use assumptions, thinking level (Low / Medium / High / Extra High on Neon), and how much of the week you actually sit on the cap all move the result. Ultra also sells Carbon access; the 5× / 20× figure is a compute multiple of Pro, not a Neon-only token pool.
These numbers are user calculations under hypothetical continuous use. They are not Google figures. Do not paste them onto a checkout page as a guarantee.
Why effective tokens vary
Three community-reported effects, all dated November 2026, sit between the theoretical month and what a heavy user actually burns.
Weekly ceiling in heavy agent use
The 5-hour refresh does not mint a new month. It refills until the weekly cap. A Pro user, @shaoxiangdashen on 20 November 2026, reported burning the entire weekly limit in about 18 hours of single-window gemini-4-neon-high agent use.
That is one person’s run, one model, one thinking level, one window. It shows the weekly ceiling can land in well under a week of concentrated agent work. After that you wait for the weekly cycle, drop to a lighter model, or look at the next official tier. See quota limits explained.
Cache drain on Gemini 4 Neon
In November 2026, multiple users reported Gemini 4 Neon prompt-cache hit rates hovering near 9% after about one minute, against roughly 95% for Opus-class caching over a longer window. @shneural (19 November) put the Antigravity effect at about 3× faster quota drain than expected. A follow-up dump circulating the same week described hit rates collapsing as the gap between turns grew: high when turns were a few seconds apart, around 9% after 1–2 minutes.
On 20 November 2026, Patrick Loeber (DeepMind, staff account) listed low cache hit rates among the feedback patterns on Neon and said the team was going back to work. Treat this as a dated, developing story, checked 20 November 2026. Cached input on the API list is $0.15 per million — a 95% discount off $3 — so a cold cache is expensive in both money (API) and allowance (subscription).
Quota buckets
Users report that the Gemini web/app allowance and the Antigravity IDE allowance are separate quota buckets. Google’s US plan card lists Antigravity rate limits as their own line (entry / higher / highest by tier) next to Gemini app usage. That split is user-reported here. Confirm on your own Usage Limits page before you assume a quiet Gemini tab refills the IDE.
Subscription vs API math
Gemini 4 Neon on the Developer API, checked 20 November 2026 at AI Studio pricing: $3 per 1 million input tokens / $15 per 1 million output tokens. Cached input $0.15. Neon is not on the Free API tier. Output includes thinking tokens.
What the monthly fee buys on that API list, versus the community theoretical monthly allowance:
| Plan | Monthly fee (US) | That fee as Neon input at $3 / 1M | That fee as Neon output at $15 / 1M | Community theoretical month (Neon) |
|---|---|---|---|---|
| AI Pro | $19.99 | ~6.7 million tokens | ~1.3 million tokens | ~800 million |
| Ultra 5x | $99.99 | ~33 million tokens | ~6.7 million tokens | ~5 billion |
| Ultra 20x | $199.99 | ~67 million tokens | ~13 million tokens | ~19 billion |
If you actually burned the community theoretical month at API list, the invoice would be in the thousands to tens of thousands of dollars even on input-only ($3 × 800M = $2,400 on Pro; $15,000 on Ultra 5x; $57,000 on Ultra 20x). Output-heavy turns cost more. An 80% input / 20% output mix, used only as an illustration, is $5.40 per million tokens — about $4,300 of API list for Pro’s 800 million.
The subscription is the heavy-user deal only if you actually burn the allowance. Light chat leaves most of that theoretical pile on the table. The direct API is for scripts and overflow after you already have a plan. It is the expensive meter for personal interactive use. Full argument: subscription vs API. Token list prices: official API token rates.
This page does not pick a plan to buy. It maps the official meter and the community conversion.
Stale prompt counts. Pre-May-2026 fixed prompt counts — Free 5/day, Pro 100/day, Ultra 500/day from 2.5-era documentation — are obsolete for personal accounts. Articles and third-party summaries that still cite them are stale. The current official page publishes compute multipliers and 5-hour windows, not those daily prompt numbers.
FAQ
Does Google publish how many tokens Pro or Ultra includes? No. The help page and the US subscriptions page publish multipliers, context windows, and a 5-hour refresh inside a weekly cap.
Are 800M / 5B / 19B official? No. They are one community conversion for hypothetical all-Neon use, posted 20 November 2026. Status partial.
Why does the weekly cap arrive in a day of agent work? Higher thinking levels, long chats, and a cold prompt cache all burn compute faster. The 5-hour refresh still stops at the weekly ceiling.
Should I skip the plan and pay the API? For personal chat and coding, buy the official plan first. Use the API for scripts and overflow. See subscription vs API.
Last checked 2026-11-20 against the links below. Open the vendor page in your region before you pay — list prices move.
- Official · Gemini Apps limits & feature availability · Open source · Checked 20 Nov 2026. Compute-based. Limit refreshes every 5 hours until the weekly cap. Plus 2× standard, Pro 4× standard, Ultra 5× or 20× Pro. Context: Free 32k, Plus 128k, Pro & Ultra 1M. Limits may change without notice.
- Official · Gemini subscriptions (United States) · Open source · Checked 20 Nov 2026. US personal: Plus $4.99, Pro $19.99, Ultra $99.99 (5× Pro) or $199.99 (20× Pro). App usage, not the Developer API invoice.
- Official · Gemini Developer API pricing · Open source · Checked 20 Nov 2026. Gemini 4 Neon Standard $3 input / $15 output per 1M tokens. Cached input $0.15. No Free tier on Neon.
- User report · @ohayou792 on X · Open source · 20 Nov 2026. Community estimate: running Gemini 4 Neon all month ≈ 800M (Pro), ~5B (Ultra 5x), ~19B (Ultra 20x). Hypothetical continuous use.
- User report · @shaoxiangdashen on X · Open source · 20 Nov 2026. Pro user reported burning the weekly limit in ~18 hours of single-window gemini-4-neon-high use.
- User report · @shneural on X · Open source · 19 Nov 2026. Reported Neon prompt-cache hit ~9% after 1 minute vs ~95% Opus-class caching, with Antigravity quota draining ~3× faster.
- Independent test · Patrick Loeber on X · Open source · 20 Nov 2026. DeepMind staff account. Listed low cache hit rates among Gemini 4 Neon feedback patterns and said the team was going back to work.