Gemini CLI's quota now comes from a paid relationship: a Gemini Code Assist licence, which Google caps at 1,500 requests per user per day on Standard and 2,000 on Enterprise, or a metered key — a Gemini API key, or a Gemini Enterprise Agent Platform key — for which Google's rate-limits page publishes no daily request count at all, only per-model limits you read inside AI Studio. The source most answers still quote — signing in with a personal Google account on the free tier, Google AI Pro or Google AI Ultra — stopped serving Gemini CLI on June 18, 2026.
That date is why this question is harder than it looks. Google's own repository README, fetched from branch main on September 21, 2026, still advertises "Free tier: 60 requests/min and 1,000 requests/day with personal Google account". It is not a mistake anyone has corrected, and it is the number most third-party articles are copying.
Where your Gemini CLI quota comes from today
| Authentication path | Daily allowance | Status on September 21, 2026 |
|---|---|---|
| Sign in with Google — Code Assist for individuals, Google AI Pro, Google AI Ultra | Formerly 1,000 / 1,500 / 2,000 requests | Stopped serving June 18, 2026 |
| Sign in with Google — Code Assist Standard licence | 1,500 requests per user per day | Live |
| Sign in with Google — Code Assist Enterprise licence | 2,000 requests per user per day | Live |
| Gemini API key | Metered; Google's rate-limits page publishes no daily request count, and per-model limits are readable only in AI Studio | Live on a paid key; unpaid keys unresolved (below) |
| Gemini Enterprise Agent Platform key (formerly Vertex AI) | Express mode publishes "Dynamic" requests per minute for the Gemini 3.x preview models | Live; express mode is free for 90 days, new Google Cloud users only |
Two first-party sources establish the cutoff. Google's announcement of May 19, 2026: "On June 18, 2026, Gemini CLI and Gemini Code Assist IDE extensions will stop serving requests for Google AI Pro and Ultra, as well as those using it free of charge using Gemini Code Assist for individuals." And the deprecation page, whose own footer reads last updated 2026-09-02: "This also applies to usage of Gemini CLI. As part of the deprecation, you can no longer use the Login with Google option to access the IDE extensions or Gemini CLI." Its FAQ answers the Standard/Enterprise question directly — those subscriptions are unaffected.
The live figures come from Google's Code Assist quota page (last updated 2026-09-18), and the mechanic on it matters more than the numbers. Quotas for agent mode and Gemini CLI are combined; one prompt can produce several model requests; the daily limits are "aggregated across all interactions with any model version or family"; and once the maximum is reached, "no further requests can be made through these interfaces to any model until the quota resets". So switching to a smaller model does not buy you extra headroom on this path, and a 1,500 figure is not 1,500 prompts. Google does not say when that reset happens — the midnight-Pacific rule you may have read is documented for the Gemini API, not for the licence path, and transplanting it is a guess.
One more trap on that same page: its "2 requests per second", 6,000-per-day and 960-per-day rows are Gemini for Google Cloud quotas for code completion and Cloud Assist panel chat. They are not the CLI's agent-mode budget, and a number lifted across is wrong even though both numbers are real.
Four official Google pages that will give you the wrong number
| Page | What it still says | Why it misleads |
|---|---|---|
| Repository README | "Free tier: 60 requests/min and 1,000 requests/day with personal Google account" | Describes the sign-in route shut off on 2026-06-18; it links the Code Assist quota page, which no longer supports the claim |
| geminicli.com quota doc | A table with Code Assist (Individual) 1,000, Google AI Pro 1,500, Google AI Ultra 2,000, Gemini API key free tier 250 | Its own footer reads "Last updated: Jun 18, 2026" — frozen on the day of the cutover |
| geminicli.com plans page | Free, Google AI Pro and Google AI Ultra merchandised as ways to run Gemini CLI | Sits directly under a site-wide banner saying those users were moved off the CLI |
| codeassist.google | "at no cost", Gemini 2.5 "with Gemini 3 coming soon", 6,000 code requests daily | Predates both the Gemini 3 line and the deprecation; the 6,000 figure belongs to a different quota system |
geminicli.com is not a squatter site — it is the repository's own declared homepage, which is exactly why it is dangerous. Discard it as out of date, not as unofficial. Its banner currently contradicts its own body copy: "Unpaid tier and Google One users: Gemini CLI was replaced by Antigravity CLI on June 18th, 2026." When the banner and the table disagree, the banner and the dated deprecation page are the later word.
One question no first-party page answers. On the API-key path, Google's May 19 blog says the CLI stays reachable via paid API keys. The maintainers' June 18 shutdown post adds only that "Enterprise users with Gemini Code Assist licenses and API key authentication remain completely unaffected" — a sentence whose subject is enterprise users, so it is not a statement that an unpaid key still works. Nothing dated after June 18 addresses unpaid keys, and the 250-requests-per-day unpaid figure survives only on the frozen quota doc. Treat an unpaid key as unverified rather than as a budget.
Which path am I on, and what does the limit look like?
Before changing anything, confirm the billing relationship. Gemini CLI documents /auth as "a dialog that lets you change the authentication method" and /stats model as showing "model-specific usage statistics, including token counts and quota information"; the quota documentation adds that a usage summary also appears on exit at the end of a session. Both are documented behaviour read on September 21, 2026, not behaviour tested here.
Two different failures produce similar-looking interruptions, and they have different remedies. A quota hit is documented as offering a named choice: "When you encounter that limit, you'll be given the option to switch to Gemini 2.5 Pro, upgrade for higher limits, or stop. You'll also be told when your usage limit resets", with a further prompt to fall back to Gemini 2.5 Flash when the 2.5 Pro limit is reached. A capacity error is model overload instead: the documentation says the CLI asks whether you want to keep trying Gemini 3 Pro or fall back to Gemini 2.5 Pro, and that "Keep trying" uses exponential backoff. Waiting fixes the second; only a reset, a licence or a different billing path fixes the first.
There is also a fallback that happens without asking. The documentation states that when the CLI detects rate limiting on the default "pro" model it "automatically switches to the 'flash' model for the current session", and that internal utility calls using gemini-2.5-flash-lite "silently fall back" to other models when quota is exhausted without changing the configured model. On a metered key that matters for the bill as well as the output: which model answered is not always the model you selected.
The four moves at the limit, with what each costs
| Line item | Published price | Source, checked September 21, 2026 |
|---|---|---|
| Code Assist Standard, monthly commitment | $22.80 per licence / month | Google Cloud pricing (also listed as $0.031232877 / hour) |
| Code Assist Standard, 12-month commitment | $19.00 per licence / month | Same table; all subscriptions billed monthly |
| Code Assist Enterprise, monthly commitment | $54.00 per licence / month | The edition whose feature bullet names increased Gemini CLI daily limits |
| Code Assist Enterprise, 12-month commitment | $45.00 per licence / month | Same table |
| Gemini API, gemini-3.1-pro-preview | $2.00 in / $12.00 out per 1M (prompts ≤200k); Free Tier "Not available" | Gemini API pricing, page last updated 2026-09-16 |
| Gemini API, gemini-3.8-flash | $0.75 in / $3.75 out through December 31, 2026; $1.50 / $7.50 from January 1, 2027 | Same page; the same dated step applies to 3.7 and 3.6 Flash |
| Gemini API, gemini-3.5-flash | $1.50 in / $9.00 out per 1M, flat | Same page — no introductory discount on this one |
| Gemini API, gemini-2.5-pro / gemini-2.5-flash | $1.25 / $10.00 (prompts ≤200k) and $0.30 / $2.50 per 1M | Same page; these are the CLI's documented fallback models |
The Google Cloud pricing table's own header reads "Gemini license type / License fees (monthly commitment) / License fees (12-month commitment)" and does not print the words "per user" next to the figure — the quota is explicitly per user per day, so read the fee as per licence until you confirm seat terms in your own order.
Move one, wait. Free, and correct when the interruption is a capacity error rather than a quota hit. Move two, buy a licence. This is the only move that raises the daily request allowance on the sign-in path, and it is the one Google documents for Gemini CLI. Move three, move to a metered key. This removes the daily request ceiling and replaces it with per-token billing plus a spend cap: Google's rate-limits page publishes spend-based limits of $10, $50 and $200 per rolling ten minutes for Tiers 1 to 3, returning 429 RESOURCE_EXHAUSTED when exceeded, and notes that limits "are not guaranteed" and per-model figures are visible only in AI Studio. The billing page (last updated 2026-09-20) also records a move from Postpay to Prepay for Gemini API usage, with a $5 minimum credit purchase. The Agent Platform variant of the same move has its own free window: express mode (last updated 2026-09-18) lets new Google Cloud users try it for up to 90 days without providing billing information, and states that the 90-day free tier is removed for existing Cloud users or once billing is enabled.
Move four, change client. Google's own recommendation is Antigravity CLI, and it is worth knowing what that costs in flexibility before you take it: Antigravity's plans documentation states there is "currently no support for: Bring-your-own-key or bring-your-own-endpoint for additional rate limits"; users not on AI Pro or Ultra are promised only "Meaningful quota, refreshed weekly" and a "Weekly rate limit", with no number printed on that page; and paid quota is "refreshed every five hours" — on Pro, "until weekly limit reached" — with limits Google says are "correlated with the amount of work done by the agent" rather than with prompt count. No Antigravity CLI binary was installed for this page, and no speed or compatibility comparison is claimed.
What does not raise a Gemini CLI quota: a Google AI Plus subscription at $4.99/month or a Google AI Pro subscription at $19.99/month (gemini.google/subscriptions, checked September 21, 2026). Since June 18, 2026 those buy Gemini app and Antigravity limits. The $100 and $200 AI Ultra tiers come from Google's May 19, 2026 announcement rather than from the subscriptions page, and carry the same caveat.
What a custom endpoint can and cannot do
Gemini CLI does document a base-URL override, and it is narrower than it looks. GOOGLE_GEMINI_BASE_URL overrides the base URL for Gemini API requests "when using gemini-api-key authentication", must be HTTPS unless it points at localhost, and has a separate sibling, GOOGLE_VERTEX_BASE_URL, for the Vertex path — which Google has since renamed the Gemini Enterprise Agent Platform while the CLI kept the old variable names. In the released v0.60.0 source, setting that variable alone selects an auth type that appears in no published documentation (GATEWAY = 'gateway'), whose key falls back to an empty string. Read that as code, not as a supported feature; it can move without a changelog entry.
Three boundaries follow, and they decide whether this is a quota answer at all. The base URL is passed as httpOptions.baseUrl into the GoogleGenAI SDK, so every request is still Google's native protocol, authenticated by default with an x-goog-api-key header; the auth enum contains no OpenAI-compatible or Anthropic provider. Kunavo's API is OpenAI-compatible and Anthropic Messages only — its route inventory at this commit has /v1/chat/completions, /v1/responses and /v1/messages and no Gemini-native generateContent path — so it is not a drop-in target for that variable, and no page here should tell you otherwise. That is an inference from protocol shape plus route inventory, not a runtime test. And the override exists only on the API-key path, so pointing at any gateway never raises, extends or resets a Google-side daily allowance; it swaps you onto a different billing relationship whose quota is then whatever that account has.
The honest version of the Kunavo option is therefore: change the client, not the endpoint. Gemini CLI vs Claude Code compares the two agents, and the Gemini API key doc covers how Gemini models are reached on an OpenAI-compatible path instead. Kunavo has runtime-tested none of these clients against its endpoint; a published setup guide is a configuration reference, not a compatibility test.
A worked estimate if you change client
These are illustrative token arithmetic, not measured task costs and not a bill ceiling. Assume one agent session that sends 300,000 uncached input tokens and receives 20,000 output tokens. Rates are live Kunavo catalog prices per million tokens; compare them against the Google list prices in the table above.
| Model | Input / output per 1M | Estimate, one session | Estimate, ten sessions |
|---|---|---|---|
| Gemini 3.1 Pro | $0.70 / $4.20 | $0.294 | $2.94 |
| Gemini 3.8 Flash | $0.525 / $2.625 | $0.210 | $2.10 |
| Gemini 2.5 Flash | $0.09 / $0.75 | $0.042 | $0.42 |
Resist converting between the two worlds. A Code Assist licence meters requests per user per day, a token key meters tokens, and Google itself says one prompt may produce several model requests — so there is no exchange rate between $22.80 a month and a per-session figure. The only honest comparison is to run the same bounded task on both and read each account's own record. Kunavo's catalog amount is a billing floor rather than a cap: when the upstream reports its charge, the bill is the greater of catalog cost and upstream cost times the applicable markup. Cache charges and external tools sit outside this example, and the minimum top-up is $10 in prepaid credit — a funding minimum, not a task fee or a subscription. See billing details.
Which route wins when
| Route | Wins when | What you give up |
|---|---|---|
| Code Assist licence, direct from Google | You are on a Workspace or Cloud account and want the sign-in path, Google's own backend features and a predictable per-seat line | A hard daily request ceiling shared across agent mode and the CLI, with no base-URL override available on that path |
| Gemini API key, direct from Google | You want one vendor's models with their own context caching and batch pricing, and metered billing instead of a daily cap | Spend caps on a rolling ten-minute window, per-model limits you can only read in AI Studio, and a January 1, 2027 price step on several Flash ids |
| An OpenAI-compatible gateway | You switch models across vendors and want one key and one balance | Not reachable from Gemini CLI itself — the protocol boundary above — so this means running a different client |
| A consumer subscription | Flat-rate use of the Gemini app or Antigravity suits you | It buys no Gemini CLI quota at all since June 18, 2026, and Antigravity's plans page prints no numeric limit and documents no bring-your-own-key or bring-your-own-endpoint for additional rate limits |
| A local model | Small or private work with no per-request charge | A capability gap against hosted frontier models, the hardware to run one, and no path into Gemini CLI's own quota system |
If you are moving the work to a different agent rather than a different Google plan, start with the OpenCode integration or the Cline setup, compare rates on the model catalog, and create a Kunavo account when you are ready to fund a key. For the wider Gemini price picture see Gemini API pricing, and for the same free-tier-that-vanished question on another agent, Codex usage limits.
FAQ
What is the Gemini CLI quota in 2026?
Google's live quota table for Gemini CLI lists two figures: 1,500 maximum requests per user per day on a Gemini Code Assist Standard licence and 2,000 on Enterprise. There is no individuals row on that page any more, and the word 'individual' does not appear on it (docs.cloud.google.com/gemini/docs/quotas, page last updated 2026-09-18, checked September 21, 2026). The other live paths are a Gemini API key or a Gemini Enterprise Agent Platform key, metered per token and per project rather than by a published daily request count for the CLI. Anyone quoting 1,000 requests per day for a personal Google account is quoting the repository README, which still advertises that allowance for a sign-in route that stopped serving consumer accounts on June 18, 2026.
Does Gemini CLI still have a free tier?
Not on a personal Google account. Google's deprecation notice states that from June 18, 2026 the Gemini Code Assist for individuals, Google AI Pro and Google AI Ultra tiers stopped being served, that this also applies to Gemini CLI, and that 'you can no longer use the Login with Google option to access the IDE extensions or Gemini CLI' (developers.google.com, page last updated 2026-09-02). Whether an unpaid Gemini API key still works inside the CLI is answered by no first-party page: Google's May 19, 2026 announcement says the CLI 'will remain accessible via paid Gemini and Gemini Enterprise Agent Platform API keys', and the maintainers' June 18 shutdown post (github.com/google-gemini/gemini-cli discussion 28017) says only that 'Enterprise users with Gemini Code Assist licenses and API key authentication remain completely unaffected' — a sentence whose subject is enterprise users, not a statement about an unpaid key. This page ran no test that would settle it, so do not budget on an unpaid key working.
How do I check my Gemini CLI quota?
Gemini CLI documents two slash commands for this. '/stats model' shows model-specific usage statistics, including token counts and quota information, and the quota documentation adds that a summary of model usage is also presented on exit at the end of a session. '/auth' opens a dialog that lets you change the authentication method, which is how you confirm which billing relationship you are actually on — a licence, an API key, or a sign-in that no longer serves your account. Both are documented behaviour on geminicli.com as read on September 21, 2026, not behaviour runtime-tested here.
How do I increase my Gemini CLI quota?
There are three paid moves and they buy different things. A Gemini Code Assist Standard licence lists at $22.80 per licence per month on a monthly commitment or $19.00 on a 12-month commitment, and carries the 1,500 requests per user per day figure. Enterprise lists at $54.00 and $45.00 on the same terms, and Google's own feature bullet for it reads 'Increased agent usage: Do more with Gemini Code Assist agent mode and the Gemini CLI with increased daily usage limits' (cloud.google.com/products/gemini/pricing, checked September 21, 2026). A metered Gemini API key replaces the daily request ceiling with a per-token bill and spend-based caps. What does not raise a Gemini CLI quota is a Google AI Plus or Google AI Pro consumer subscription: since June 18, 2026 those apply to the Gemini app and to Antigravity, not to Gemini CLI.
Why did Gemini CLI stop working after I signed in with Google?
Because eligibility is decided on Google's side, not by your client version. The released v0.60.0 source comments that for the free tier, 'Eligibility is determined by the Code Assist server response', and that the CLI 'Surfaces ineligibility reasons for the FREE tier from the server'. So an up-to-date binary still offers 'Sign in with Google' in its auth dialog even for an account the server will refuse, which is why the failure looks like a client bug rather than a policy change. The exact refusal text a consumer account sees today was not captured for this page.
Can I point Gemini CLI at a different API endpoint or a gateway?
Only at a Gemini-native one, and only on the API-key path. Gemini CLI documents GOOGLE_GEMINI_BASE_URL as overriding the base URL for Gemini API requests when using gemini-api-key authentication, HTTPS-only unless it points at localhost; Vertex, now the Gemini Enterprise Agent Platform, has a separate GOOGLE_VERTEX_BASE_URL. In the released v0.60.0 source that URL is passed as httpOptions.baseUrl into the GoogleGenAI SDK, so every request is still Google's native protocol, authenticated by default with an x-goog-api-key header, and the auth enum contains no OpenAI-compatible or Anthropic provider. Kunavo's API is OpenAI-compatible plus Anthropic Messages and has no Gemini-native generateContent route, so it is not a drop-in target for that variable — the honest route is to change the client rather than the endpoint. Note also that the override exists only on the API-key path, so it never raises, extends or resets a Google-side daily allowance.
Checked September 21, 2026: Google's deprecation notice, the Code Assist quota page, the Google Cloud pricing table, the Gemini API pricing, billing and rate-limits pages, the geminicli.com documentation corpus and its quota page, the Antigravity plans page, the repository README, the GitHub repository API and the npm registry (stable 0.60.0, published 2026-09-15). Not checked: no Gemini CLI or Antigravity CLI binary was installed, no Google account was signed in, and no proxy was pointed at the CLI — the protocol boundary is read from the released v0.60.0 source and from Kunavo's own route inventory. Kunavo token rates come from the live catalog and every dollar figure here is illustrative token arithmetic.