Gemini CLI free tier: request limits, models, and what happens to your code
The Gemini CLI free tier on a personal Google account: per-minute and per-day request limits, what happens at the limit, how your input is handled, and when to move to an API key.
Contents
The appeal of Gemini CLI is that signing in with a personal Google account costs nothing. That free tier comes with two conditions worth knowing before you point it at anything important: a cap on requests, and the possibility that your input is used for product improvement.
This article covers what the free tier includes, what happens at the limit, how data is handled, and when to move to a paid path. The numbers change, so treat the official documentation as the current source.
KEY POINT
What you will learn
- The free tier's request limits and which models you get
- When your input may be used for product improvement, and the alternatives
- How to decide between the free tier, an API key and Vertex AI
Three authentication paths
| Path | Cost | Limits | Data handling |
|---|---|---|---|
| Personal Google account (free) | Free | Capped requests per minute and per day | May be used for product improvement; sometimes controllable in settings |
| Gemini API key (paid) | Per-token usage billing | Depends on your tier | Paid API terms state input isn't used for training |
| Vertex AI (Google Cloud) | Per-token usage billing | Your project's quota | Governed by Google Cloud's data terms, aimed at organizations |
A Google Workspace account or a Gemini Code Assist license is handled under the terms of that agreement instead.
The free tier's limits
Officially there is a cap on requests per minute and on requests per day. As of this article's verification date (September 7, 2026), the README showed 60 requests per minute and 1,000 per day for a personal account — figures that do get revised, so check the README rather than relying on these.
The important part is how fast they go. Gemini CLI runs as an agent, so one instruction from you turns into several model calls. "1,000 per day" is not 1,000 prompts; in practice it is a fraction of that. Run a few long tasks and you can reach the cap by the afternoon.
用語解説
Requests versus turns: within a single turn — one instruction from you — Gemini reads files, runs commands and checks results, calling the model each time. The limits count model calls, not turns.
Which models you get
The free tier gives you the current top Gemini model available at sign-in. Under load, or as you approach the limit, you may be switched to a lighter model automatically. /about and /stats show what the session is currently using.
Model names change with each release, so this article doesn't fix one. The shape to remember is two tiers — a capable model and a lighter one — with switching between them depending on conditions.
What happens at the limit
- Requests fail with a rate-limit error
- You are switched to a lighter model and answers get noticeably weaker
- The limits reset on their own schedule, per minute and per day
The quickest way to keep working is an API key. Set GEMINI_API_KEY and restart, and the session runs on the paid tier from then on.
export GEMINI_API_KEY="your-key"
geminiHow your input is handled
Free use on a personal Google account comes with notice that your prompts and code may be used for product improvement. Your Google account settings may let you turn off usage sharing, but the details of the terms get revised — read the official explanation rather than assuming.
For work code or customer data, the rules of thumb:
- Keep the free tier for personal learning and evaluation
- For work, use a paid API key or Vertex AI, and read the data terms
- Confirm your organization allows sending code to an external AI service at all
- Configure the agent not to read secrets — see Keeping secrets out of your AI coding tools
Free is not a reason to feed it work code
The free tier's data terms are not the paid API's data terms. It's the same check worth doing on a Claude Code personal plan: confirm the conditions for your authentication path before using it for work. The Claude Code side is covered in Is your code used to train Claude?
When to move to a paid path
| Situation | Recommendation |
|---|---|
| Personal learning, small open-source fixes | The free tier is enough |
| Hours of use daily, hitting the cap often | An API key, billed per token; watch usage in Google AI Studio |
| Work code | An API key or Vertex AI, after reading the data terms |
| Organization-managed use, audit requirements | Vertex AI |
API key cost follows input and output tokens. For ordinary development it usually lands in the low single-digit dollars per month to low tens. Check /stats for token usage and estimate from your own numbers rather than a guess.
Summary
- The free tier runs on a personal Google account with per-minute and per-day request caps, and one instruction spends several requests
- At the limit you get errors or a lighter model; setting
GEMINI_API_KEYmoves you to the paid tier - On the free tier your input may be used for product improvement — use an API key or Vertex AI for work code
- Check the official documentation for current limits and data terms rather than any article's numbers
FAQ
- What are the free tier limits?
- Signing in with a personal Google account caps requests per minute and per day. The figures in the official README get revised, so check there for current numbers.
- Is my code used for training on the free tier?
- Free use on a personal account comes with notice that data may be used for product improvement. For work code, use a paid API key or Vertex AI, where the terms cover how data is handled.
- What happens when I hit the limit?
- Requests fail with a rate-limit error, or you may be switched to a lighter model automatically. Wait for the reset, or switch to an API key.
Primary sources
This article was drafted by AI from official documentation and reviewed by the site operator before publishing. Found a mistake? Let us know via the contact page.