Save money

Gemini CLI's free tier is gone: what still works for free

Q: What replaced Gemini CLI for free users?

Antigravity CLI, a separate binary called agy. It reads your existing GEMINI.md and AGENTS.md files, imports old extensions with agy plugin import gemini, and on the free plan runs on a weekly quota whose size Google does not publish.

Google stopped serving Gemini CLI to free, AI Pro and Ultra users on June 18, 2026. Who kept access, what is still free, and how to keep using Gemini.

Kodelyth · The team behind kRouter

Oct 5, 2026 · Updated Oct 8, 2026

8 min read

ShareX LinkedIn Hacker News Reddit

You run gemini, sign in with your Google account the way you have since 2025, and it no longer gets you anywhere. Gemini CLI's documentation pages now open with a banner: for unpaid tier and Google One users, Gemini CLI was replaced by Antigravity CLI on June 18, 2026.

Nothing is wrong with your install, your token or your network. Google stopped serving that sign-in, so logging in again, reinstalling or trying another personal account all end the same way.

Below: who lost access, who kept it, what is still free in October 2026, and where a router helps and where it does not.

What changed on June 18

Google announced the change on May 19, 2026, and made Antigravity CLI available to everyone the same day. From June 18:

Gemini CLI and the Gemini Code Assist IDE extensions stopped serving requests for Gemini Code Assist for individuals (the free Google sign-in), Google AI Pro and Google AI Ultra.
Gemini Code Assist for GitHub stopped taking new installations on GitHub organizations, and Google said requests would stop being served in the following weeks.
Google points all of them to Antigravity CLI and the Antigravity desktop app, which share one agent harness.

Paying did not protect you: Pro and Ultra subscribers lost Gemini CLI along with free users.

Who still has Gemini CLI

Google's announcement names the users whose access did not change:

Organizations with a Gemini Code Assist Standard or Enterprise license, or using Code Assist for GitHub through Google Cloud.
Anyone using a paid Gemini API key.
Anyone using a paid Gemini Enterprise Agent Platform key. That is the name Google gave Vertex AI in April 2026.

The open question is an unpaid key. Gemini CLI's quota page still lists a free Gemini API key tier (250 requests a day, Flash models only) and Vertex AI express mode, but the same page carries the replacement banner, and the announcement only promises access with paid keys. Until Google says otherwise, plan on a paid key if you want to stay on Gemini CLI.

What does not work

Signing in again. This is not an expired token; the sign-in itself no longer grants access.

Reinstalling, or pinning an older Gemini CLI. The cutoff is on Google's side, and an older client calls the same place.

Switching to the Gemini Code Assist IDE extension. It stopped serving the same users on the same day.

Upgrading to Google AI Pro or Ultra. Both lost Gemini CLI too. A subscription now buys Antigravity limits.

Another personal Google account. Same tier, same refusal.

Sending the old sign-in through a proxy. A tool that reuses Gemini CLI's Google login talks to the same backend, so expect the same cutoff. That includes kRouter's own Gemini CLI provider, covered below.

What is still free

Antigravity CLI's free plan

Google's replacement is a separate binary, agy. The first time you run it on a machine with a Gemini CLI setup, it offers a checklist of what to migrate. It still reads GEMINI.md and AGENTS.md, and agy plugin import gemini converts old extensions into Antigravity plugins. Two things need attention (migration guide):

Workspace skills have to be moved from .gemini/skills/ to .agents/skills/ by hand.
MCP servers move out of ~/.gemini/settings.json into mcp_config.json files (~/.gemini/config/mcp_config.json globally, .agents/mcp_config.json per workspace), and a remote server's url or httpUrl key becomes serverUrl.

Google publishes no numbers for its limits. Its plans page gives users without Pro or Ultra "a meaningful quota, refreshed weekly", and says usage tracks how much work the agent does, so complex tasks drain it faster. Pro and Ultra refresh every five hours, up to weekly limits, and only they can buy AI credits beyond the quota. Every plan gets Gemini 3.1 Pro and Gemini 3.8 Flash. Bring-your-own-key and bring-your-own-endpoint are not supported, so you cannot add capacity of your own.

A Gemini API key from AI Studio

The Gemini API still has a free tier, and it is the most flexible free Gemini access left. On Google's pricing page in early October 2026, the free tier covers Gemini 3.5 to 3.8 Flash, the 3.5 and 3.1 Flash-Lite models, Gemini 3 Flash Preview, and Gemini 2.5 Pro, Flash and Flash-Lite. Gemini 3.1 Pro Preview has no free tier.

Before you lean on it:

Limits are per project, not per key. Your numbers are shown in AI Studio, they vary by model, and daily quotas reset at midnight Pacific time (rate limits). A second key in the same project adds nothing.
Free-tier content is used to improve Google's products. Keep code you are not allowed to share off it.
An agent spends it quickly. A coding agent sends far more requests per minute than a chat does, so expect 429s on a free project (how to read them).

Express mode, if you are new to Google Cloud

Express mode gives a @gmail.com account that is new to Google Cloud an API key and up to 90 days of free use within its quotas, with no billing details. Its model list includes Gemini 3.1 Pro Preview -- the model the Gemini API free tier leaves out -- with limits Google describes as dynamic, while Gemini 2.5 Pro and 2.5 Flash are capped at 10 requests a minute. Existing Google Cloud users do not get the free 90 days, and express mode is still a pre-GA preview.

The structural fix: keep your agent, change where Gemini comes from

Antigravity's free quota is meant for Google's own CLI and app. The free Gemini your own agent can use without risking an account is API access: an AI Studio key, and an express-mode key if you qualify. A local router lets the agent you already use draw on several of these sources, moving to the next when one runs out.

That is what kRouter does. Install it and start it in the background:

npm install -g @sifxprime/krouter
krouter -t

Then, in the dashboard at http://localhost:20128/dashboard:

Providers → Gemini, and paste your AI Studio key. The provider page can list the models Google currently serves to that key under From API, so a new Flash release can be added as soon as Google lists it.
If you have an express-mode key, Providers → Vertex AI, and paste it. The provider accepts a plain API key as well as a service account JSON. Send one test request before you rely on it.
Combos → Create Combo, and add models in order of preference:

Combo "gemini-free"
  1. vx/gemini-3.1-pro-preview      express-mode key, first 90 days
  2. gemini/gemini-flash-latest     Gemini API, free-tier key
  3. gemini/gemini-2.5-pro          same key, with its own per-model limits

CLI Tools: open the card for your agent -- Claude Code, Codex, OpenCode and others are there -- choose gemini-free as the model (on the Claude Code card, for each model slot you want it to cover) and click Apply. The agent now talks to http://localhost:20128/v1. From the same machine it needs no key unless you have set REQUIRE_API_KEY=true. For what the Codex card writes, see Codex CLI with any model.

gemini-flash-latest is Google's own alias for the newest Flash release, so that entry follows each new model without a config change. When an entry fails with a rate limit or quota error, the request moves to the next one.

Three caveats:

None of this adds quota. A combo spends each free allowance in turn, and when all are spent, requests fail until the next reset. A no-account option such as MiMo Code Free (mmf/mimo-auto) can go last if you would rather get an answer from a different model than none.
Claude Code needs the context window set. It assumes 200K for a model name it does not recognize, and a combo name is one of those. The Max context setting on the Claude Code card writes CLAUDE_CODE_MAX_CONTEXT_TOKENS. It offers 200K, 300K, 500K and 1M; pick the largest that does not exceed the smallest window in the combo (why it matters).
Antigravity can join the combo, with a risk. ag/gemini-pro-agent is Gemini 3.1 Pro through an Antigravity sign-in, which is not licensed for router use. kRouter shows a risk notice before you connect it, and Google can restrict the account (what helps). Antigravity reports its remaining quota, and once kRouter has that figure it tries the entry ahead of entries that report none, whatever position you gave it.

What happens to kRouter's Gemini CLI provider

kRouter still lists a Gemini CLI provider, marked deprecated in the dashboard and carrying a risk notice. It signs in with Gemini CLI's own Google login and calls the same Code Assist backend, so Google's cutoff applies to it as well: expect a free, AI Pro or Ultra personal account connected there to be refused just as the CLI is. If one is still in a combo, remove it; it only adds failed attempts.

Code Assist Standard and Enterprise seats kept Gemini CLI, but this provider has not been tried with one, so this post makes no claim for organizational accounts.

The reverse is not available either. Google does not support bring-your-own-endpoint in Antigravity, so there is no supported way to point Antigravity CLI at kRouter, and kRouter's Antigravity MITM tool is built for the Antigravity IDE, not the CLI.

Choosing

Your situation	What to do
Your company has Code Assist Standard or Enterprise	Nothing changed, keep Gemini CLI
You have a paid Gemini API or Agent Platform key	Keep Gemini CLI on that key
You want Google's own agent and pay nothing	Antigravity CLI on the free plan, weekly quota
You pay for Google AI Pro or Ultra	Antigravity CLI; the plan now buys its five-hour refresh
You want free Gemini models in Claude Code, Codex or OpenCode	An AI Studio key, plus an express-mode key if you are new to Google Cloud, behind a kRouter combo
You need Gemini 3.1 Pro at no cost	Express mode for its 90 days if you qualify; otherwise Antigravity's quota

Common questions

Is Gemini CLI still free?

Not for personal accounts. Since June 18, 2026, Google no longer serves it to the free Code Assist for individuals sign-in or to Google AI Pro and Ultra subscribers. It still works with a Code Assist Standard or Enterprise license, and with paid Gemini API or Gemini Enterprise Agent Platform keys.

Does a free Gemini API key still work?

For the Gemini API itself, yes: the current Flash and Flash-Lite models and Gemini 2.5 Pro have a free tier with per-project limits, but Gemini 3.1 Pro Preview does not. Whether Gemini CLI keeps accepting an unpaid key is less clear, because Google's announcement only names paid keys.

What replaced Gemini CLI for free users?

Antigravity CLI, a separate binary called agy. It reads your existing GEMINI.md and AGENTS.md files, imports old extensions with agy plugin import gemini, and on the free plan runs on a weekly quota whose size Google does not publish.

Can I use Antigravity CLI with my own API key or through a router?

Not in any supported way. Google's plans page says bring-your-own-key and bring-your-own-endpoint are not supported. Once the quota runs out, you wait for the refresh, upgrade to Pro or Ultra, or, on those plans, buy AI credits.

Can kRouter bring the free Gemini CLI tier back?

No. Only Google decides which accounts it serves, and kRouter's own Gemini CLI provider uses the same sign-in and backend. What kRouter can do is put the free Gemini access that remains -- an AI Studio key, an express-mode key -- behind the agent you already use, and move to the next source when one runs out.