Gemini CLI Pricing (2026): Plans & Real Costs

If you searched “gemini cli pricing” expecting a free tier with 1,000 requests a day, that page is out of date. On June 18, 2026, Google stopped serving Gemini CLI requests for free, Google AI Pro, and Google AI Ultra accounts, as well as individual Gemini Code Assist users, so most pricing guides that show up today describe a setup that’s now wrong.

So what does Gemini CLI actually cost in mid-2026? For individuals, the generous free CLI is gone: you either move to Google’s new Antigravity CLI on a much smaller quota, or pay through a Gemini API key. For teams, Gemini CLI still runs, but only on an enterprise Code Assist license or paid API keys. This post walks through every current price, the quotas that bite, and where the real spend hides.

Is Gemini CLI free? The free tier and its limits

For most readers asking this in 2026, the answer is no longer yes. The free Google account login path that made the Gemini CLI famous was discontinued on June 18, 2026. Google announced the change on the official repo and confirmed that Gemini CLI “will stop serving requests for Google AI Pro, Google AI Ultra, and free-tier individual accounts.”

Here is what still costs nothing, and what doesn’t:

• Antigravity CLI free tier: Individuals are pushed to Google’s new agy binary, which has its own much smaller free quota than the old Gemini CLI allowance. Google hasn’t published an exact daily request number, but community testing reports a free ceiling in the low tens of requests per day, down from the roughly 1,000 per day that the Gemini CLI used to offer.
• Gemini API free tier: The Gemini Developer API still has a free tier through Google AI Studio, where content can be used to improve Google’s products. It doesn’t authenticate Gemini CLI for individuals anymore, so it helps you build apps, not run the CLI for free.
• The Apache-2.0 repo: The Gemini CLI source is still open under Apache 2.0, so you can install it. It just won’t provide you with model responses without a paid API key or an enterprise license.

The free experience for individuals is now Antigravity CLI, on a tighter leash.

Paid plans (Google AI Pro / Ultra, Code Assist Standard/Enterprise)

The Gemini subscription names you’ve seen quoted, Google AI Pro and Google AI Ultra, no longer grant Gemini CLI access; as of June 18, those consumer plans route terminal usage through Antigravity instead. If you want Gemini CLI itself, the supported paths are enterprise licenses or API keys.

Two Gemini Code Assist licenses keep Gemini CLI working for teams, with prices verified on Google Cloud’s pricing page:

Annual commitments cut the per-seat price by roughly 17 percent on both tiers. A five-person team on the annual Standard plan pays about $95 a month for 1,500 requests per developer per day. Enterprise adds higher quotas plus the security and compliance controls larger orgs need.

Pay-as-you-go / API pricing (and when you cross into it)

If you still want real Gemini CLI, not Antigravity, a paid API key is the route: install the open-source CLI, drop in your key, and pay per token. There’s no monthly minimum; you pay for what the agent consumes.

These are the current paid-tier rates per one million tokens, from the official Gemini API pricing page:

Coding agents are token-hungry: they reread files, replay diffs, and stuff context windows on every turn, so a single feature task on a Pro-class model can chew through hundreds of thousands of tokens, and output tokens, including the model’s thinking tokens, cost several times more than input. Flash-Lite keeps experiments cheap; a Pro model on a large repo can turn a casual afternoon of vibe coding into a real bill.

Quotas and limits (requests/min, /day, the “free tier” gotchas)

Quotas, not headline prices, are what actually shape your day. The old Gemini CLI free tier was famous for being roomy, widely reported to be around 60 requests per minute and 1,000 per day on a personal Google login, a number that stale articles still cite even though it no longer applies to anyone.

What replaced it is tighter and more fragmented, as the table of plans above shows. Antigravity’s free ceiling is a fraction of the old numbers; Code Assist Standard and Enterprise cap out at 1,500 and 2,000 requests per user per day, and a paid API key is governed by your project’s per-model rate limits rather than a flat daily cap, so the practical ceiling is your budget. Antigravity’s refresh cadence reportedly rolls rather than resetting cleanly each day, so treat any exact free-tier figure as community-sourced until Google publishes one.

The “free tier is a lie” complaint flooding Reddit threads is mostly about this gap: people remember 1,000 requests a day, install Antigravity expecting the same, and burn through their weekly allowance in minutes because the new harness is more verbose and the quota is shared across models. Check the quota that applies to your plan today, not the one a one-year-old blog post promises.

Gemini CLI vs Claude Code vs Codex on real cost

Comparing coding agents on price is harder than it looks: the three big options bill differently. Claude Code and Codex are commonly used through flat subscription plans or their respective APIs on a per-token basis. Gemini CLI, post-June, leans on either per-token API billing or a per-seat Code Assist license. None of them map cleanly to “X dollars a month for unlimited coding.”

Subscription plans give predictable costs and a soft usage ceiling, good for steady daily use. Per-token billing has no ceiling and no floor, suiting bursty or automated workloads but punishing runaway agents. Gemini’s Flash-Lite rates are among the cheapest per token of the frontier-adjacent models, so high-volume light tasks can be very cheap, while heavy reasoning on a Pro model narrows the gap with competitors. We dig into model-by-model tradeoffs in our guide to the best LLM for coding and compare two popular CLIs head-to-head in Codex vs Claude Code.

Picking an agent based on this month’s price sheet is fragile. June 18 proved that a vendor can change the terms, the tool, and the quota overnight.

Hidden costs and how to control spend

The sticker price is rarely the real price. Teams are surprised less by the rate card and more by how agents behave: chatty models that explain every step burn through output tokens quickly, and long sessions on big repos multiply input tokens as the agent reloads files and history at each turn. Tool calls add up too, and search grounding on Gemini 3 models is free for the first 5,000 prompts a month, then $14 per 1,000 queries. There’s also a migration tax: when a vendor sunsets a tool, like Gemini CLI for individuals, rewriting your scripts and CI pipelines is a real cost that no pricing page shows.

Controlling spend mostly comes down to a few habits: match the model to the task so Flash-Lite handles linting and Pro handles architecture, keep sessions scoped so the agent isn’t dragging an entire monorepo into context for a one-line fix, and avoid hard vendor lock-in so a pricing change like June 18 is an annoyance, not a fire drill.

Pricing for teams: running agents in your own cloud

For a solo developer, the calculus is “which key do I paste in?” For a team, it’s a procurement problem: per-seat licenses, per-token overages, separate quotas per developer, and the standing risk that your vendor changes the deal, as June 18 just did.

This is the gap we’re built to close. We’re model-agnostic orchestration: we run Claude Code, Codex, Gemini-backed agents, and whatever ships next, across your repos and tools, so your team isn’t tied to one vendor’s pricing or quota changes, and you swap the underlying agent instead of rebuilding your pipeline if one reshuffles its CLI. Our self-hosted deployment puts the whole system on a single VM inside your own cloud account, behind your VPC, on AWS, GCP, or Azure (GCP is currently in early access), with your own API keys so model spend stays on your existing billing and your code never leaves your network. Our plans run from a free tier, 10 free credits as a one-time grant rather than a recurring refresh, with the option to top up anytime, to Pro at $60 a month and Max at $200 a month, with Enterprise adding SSO, bring-your-own-key, and the self-hosted option. This isn’t about replacing your Gemini bill; it’s about making sure no single vendor’s pricing change can hold your team’s workflow hostage.

Where to go from here

Gemini CLI pricing in 2026 is not a single number. It is a fork in the road: Antigravity CLI on a small free quota, a paid Gemini API key billed per token, or a Code Assist seat for teams. Pick based on how your usage actually looks, and re-check the quotas, because they change.

If you would rather not re-architect your workflow every time a vendor changes its CLI, start with us. Spin up our free tier to run any agent across your repos, or book a demo to see the self-hosted, bring-your-own-key setup that keeps model spend and source code inside your own cloud.

Frequently asked questions

Is Gemini CLI free? Not for individuals anymore. The free Google-login tier was discontinued on June 18, 2026. Individuals move to Antigravity CLI on a much smaller free quota or pay through a Gemini API key. The source code remains free and open under Apache 2.0, but it needs a paid key or enterprise license to serve responses.

Is Gemini CLI cheaper than Claude Code? It depends on the usage type. Gemini’s Flash-Lite per-token rates are very low, so high-volume light tasks can be cheaper on Gemini, while heavy daily reasoning work is often more predictable on a flat Claude Code or Codex subscription. Compare model tradeoffs in our agentic coding guide.

Does Gemini CLI need a subscription? To run Gemini CLI itself, you need either a paid Gemini API key or a Gemini Code Assist Standard or Enterprise license. The free Antigravity CLI is a separate tool, not the Gemini CLI.

How much can I use on the free tier? Far less than before. The old Gemini CLI offered roughly 1,000 requests per day; community reports put the new Antigravity free ceiling in the low single digits per day. Google hasn’t published an official figure, so verify against your account before relying on it.