Gemini CLI Alternatives: Antigravity vs Claude Code (2026)

If you used Gemini CLI for free, the closest replacement is Google's Antigravity CLI (agy), which has a free tier on Gemini models and reads your existing GEMINI.md files. If you want the best terminal agent for a codebase you care about, pick Claude Code on Claude Pro at $20 a month, or Codex if you already pay for ChatGPT. Gemini CLI itself still runs on a paid Gemini API key, which only makes sense for teams that need its Apache 2.0 source.

On June 18, 2026, Gemini CLI stopped answering for most of the people who used it. Google's developer blog had announced it a month earlier, on May 19: Gemini CLI and the Gemini Code Assist IDE extensions would stop serving requests for Google AI Pro and Ultra subscribers and for anyone on the free Gemini Code Assist for individuals. A lot of people found out the way I did. A script that had called gemini -p from a git hook for months started failing, and the hook was the thing that told me.

This page is for that moment. AI Tools Digest does not make or sell any of these tools, and none of them paid to be here. Facts below come from Google's announcement and the Antigravity docs, and from Anthropic's and OpenAI's plan pages as of November 2026.

Gemini CLI alternatives compared

Tool	Best for	Free tier	Starting paid price	Project instruction file	Key limit
Antigravity CLI (agy)	Former free Gemini CLI users, Google AI Pro subscribers	Yes, a quota on Gemini models that refreshes weekly	Included with Google AI Pro, $19.99 a month	GEMINI.md and AGENTS.md	No published source; third-party models are Ultra-only
Claude Code	Careful work on production code, teams that want hooks	No; the free Claude plan does not include it	Claude Pro, $20 a month or $17 billed yearly	CLAUDE.md; AGENTS.md only when no CLAUDE.md exists	Five-hour and weekly limits, shared with the Claude chat apps
Codex CLI	ChatGPT subscribers, handing off well-scoped tasks	No; not on the free ChatGPT plan	ChatGPT Go, $8 a month, for lightweight tasks	AGENTS.md, capped at 32 KiB combined	OpenAI models only
Gemini CLI on a paid API key	Teams on Gemini Code Assist Standard or Enterprise, or anyone who needs the open-source code	No, for individuals, since June 18, 2026	Usage-based Gemini API billing	GEMINI.md	Its README says it is not tuned for the Gemini 4 series

Google AI Pro, Claude and ChatGPT prices are from each vendor's plan page as of November 2026. Google's Antigravity plans page lists quotas by tier but gives no prices of its own.

What actually changed on June 18

Google AI Pro and Ultra subscribers lost access, and so did everyone on the free Gemini Code Assist for individuals. Gemini Code Assist for GitHub also stopped taking new installs on individual GitHub organizations that day.

Two groups kept it. Anyone on a Gemini Code Assist Standard or Enterprise license saw no change, and Gemini CLI stays available to anyone paying through a Gemini API key or a Gemini Enterprise Agent Platform key.

The replacement Google named is Antigravity CLI. It is written in Go, and it shares its agent harness with the Antigravity 2.0 desktop app, so Agent Skills, hooks, subagents and extensions (now called plugins) work in both. Google's post admits there would not be "1:1 feature parity right out of the gate."

What did not carry over is the license. Gemini CLI is Apache 2.0, and its source is on GitHub. The public antigravity-cli repository holds a README, examples and issue templates, and when I looked I found no source directory and no license file. If you picked Gemini CLI because you could read and fork it, Antigravity does not replace that part, and the Cline, Roo Code and Aider comparison is a better place to look.

Antigravity CLI, the direct replacement

You install it with a shell script (curl -fsSL https://antigravity.google/cli/install.sh | bash on a Mac or Linux), and it lands in ~/.local/bin/agy. The first time you run agy in a machine that had Gemini CLI, it notices the old config and offers a migration checklist, which converts extensions and global settings and moves your saved Google session into the OS keychain.

The good news is the instruction files. Antigravity reads GEMINI.md and AGENTS.md in the workspace, plus your global ~/.gemini/GEMINI.md, so a repo's rules carry over untouched.

The quotas are where it differs from what you had. Google's plans page describes the tier for people with no Google AI plan as "a meaningful quota, refreshed weekly." Google AI Pro refreshes every five hours until a weekly limit, and Ultra gets the most. That weekly reset on the free tier is the trap: Gemini CLI's old free allowance was a daily request cap, so a heavy Monday used to cost you Monday. Now it can cost you the week. Pro and Ultra can switch on AI credits to keep working past the quota, billed at Gemini Enterprise rates. The free tier cannot.

Permissions are sensible by default. In the CLI's default mode it asks you to review the diff before it changes a file, and shell commands wait for approval. accept-edits lets file changes through. auto runs commands in a sandbox with no network access and judges the rest itself, though it is not available on business or enterprise accounts. The flag that turns all of this off is called --dangerously-skip-permissions, which is honest of them.

The migration step that breaks MCP servers

This is the one that cost me an afternoon.

Gemini CLI kept MCP servers inline in ~/.gemini/settings.json. Antigravity moves them to their own files: ~/.gemini/config/mcp_config.json for global servers and .agents/mcp_config.json inside a project. Remote servers also changed key names. Where Gemini CLI wanted url or httpUrl, Antigravity wants serverUrl. I had copied a remote server across by hand, kept the old key, and spent far too long assuming the server itself was down.

Workspace skills moved too. Anything in .gemini/skills/ has to go to .agents/skills/ before Antigravity will offer it as a slash command. Extensions convert with one command, agy plugin import gemini, which turns extension commands into skills and pulls their MCP entries into the new config file.

Run the import first, then open the new mcp_config.json and search it for url. That check takes a minute.

Claude Code and Codex, if you are leaving Google

Both are paid, and both have had longer to mature than Antigravity CLI. Our Claude Code vs Codex comparison covers them in depth, so here is only what matters to someone coming from Gemini CLI.

Your GEMINI.md will not be read by either one. Codex reads AGENTS.md from the git root down to your working directory. Claude Code reads CLAUDE.md, and falls back to AGENTS.md only when there is no CLAUDE.md anywhere in the way. The fix I use on every repo is to rename GEMINI.md to AGENTS.md, then add a one-line CLAUDE.md containing @AGENTS.md. Antigravity reads AGENTS.md as well, so one file now serves all three.

Claude Code asks before it edits a file or runs most shell commands. That is slower than Antigravity's accept-edits, and on a repo with a deploy hook, it is what I want. Codex starts in a workspace-write sandbox and runs routine commands inside the project without asking, stopping only for network access or files outside the workspace.

On limits, the difference is how they are pooled. Claude Code shares one five-hour and weekly budget with the Claude chat apps. Codex shares its allowance between local CLI runs and cloud tasks. Either way, the $20 plans are enough to find out whether you like the tool, and not enough for full-day agent work, which is where Claude Max at $100 and ChatGPT Pro at $100 start.

Where each option fits (and where it does not)

Antigravity CLI fits if you used Gemini CLI because it was free, or you already pay for Google AI Pro. It reads your old files and costs nothing to try. It does not fit if the open-source license mattered to you, or if you want Claude or GPT models on anything below Ultra.

Claude Code fits a production repo where you want to approve every edit and enforce rules with hooks. It falls short on price for occasional use, since there is no free way in.

Codex fits if ChatGPT is already on your card statement and you like handing off a ticket and coming back to a branch. It is weaker if you want to watch each step.

Gemini CLI on an API key is for teams already on Code Assist Standard or Enterprise. For an individual, paying per token for the tool Google has stopped building toward is the wrong bet.

My own setup since June: Claude Code for the main work on this site, and Antigravity's free tier kept installed for second opinions on Gemini, which costs nothing as long as I stay under the weekly quota. For the wider field, including editor-based agents, see the ranked list of AI coding agents. If what you miss is running a model with no account at all, Ollama, LM Studio and llama.cpp on a Mac covers that.

Frequently Asked Questions

Is Gemini CLI shut down?

Not entirely. On June 18, 2026, Gemini CLI stopped serving requests for Google AI Pro and Ultra subscribers and for free users of Gemini Code Assist for individuals. It still works with paid Gemini API keys and with Gemini Code Assist Standard or Enterprise licenses.

What replaced Gemini CLI?

Google replaced it with Antigravity CLI, a terminal agent written in Go that you run with the agy command. It has a free tier with a quota that refreshes weekly, is included in Google AI Pro and Ultra, and reads existing GEMINI.md and AGENTS.md files.

Is Antigravity CLI free?

Yes, with limits. Users without a Google AI plan get a quota on Gemini models that refreshes weekly. Google AI Pro refreshes every five hours up to a weekly limit, and Pro and Ultra users can buy AI credits to keep working after the quota runs out.

How do I move MCP servers from Gemini CLI to Antigravity?

Run agy plugin import gemini to convert extensions, then check your MCP config. Antigravity reads servers from ~/.gemini/config/mcp_config.json or .agents/mcp_config.json rather than settings.json, and remote servers must use the serverUrl key in place of url or httpUrl.

Is Claude Code or Codex a better Gemini CLI alternative?

Claude Code is better if you want to approve each edit and command on a production codebase, starting at $20 a month on Claude Pro. Codex is better if you already pay for ChatGPT and want to hand off tasks that run in a sandbox, starting at $8 a month on ChatGPT Go. Neither reads GEMINI.md, so rename it to AGENTS.md.