Last updated on
Gemini CLI alternatives in 2026: what to use instead
Gemini CLI’s consumer free tier is gone. As of June 18, 2026, Gemini CLI and the Gemini Code Assist IDE extensions stopped serving requests for free, Google AI Pro, and Google AI Ultra accounts. Google points everyone to the new Antigravity CLI — but for free users that’s a big downgrade, so weigh the full set of options before you migrate.
Here’s the short version. If you want a terminal agent with no subscription, go open source: OpenCode or Aider. If you want Google’s official path, that’s Antigravity (the IDE is more generous than the CLI right now). If you’ll pay for the most capable terminal agent, it’s a two-horse race between Claude Code and Codex (OpenAI’s).
The alternatives at a glance
| Tool | What it is | Cost | Best for |
|---|---|---|---|
| Antigravity CLI | Google's official successor (binary agy), closed source |
Free tier ~20 req/day; included in AI Pro/Ultra | Staying in Google's ecosystem |
| Antigravity (IDE) | Agent-first VS Code fork, runs multiple agents | Free during public preview | Agent-driven workflows in an editor |
| OpenCode | Open-source terminal agent, 75+ model providers | Free; bring your own API key | No subscription, model choice |
| Claude Code | Anthropic's terminal agent, full-codebase aware | Pro from $20/mo (no free tier) | Top-tier paid terminal agent |
| Codex CLI | OpenAI's terminal agent, open source (Rust) | Free CLI; ChatGPT plan or BYO API key | Top-tier OpenAI terminal agent |
| Aider | Open-source, git-native CLI pair programmer | Free; bring your own API key | Lean, git-first terminal workflow |
How capable are they?
If you’re leaving Gemini CLI, the fair question is whether you’re trading down. On raw terminal-agent ability, you’re not — the best alternatives score higher than Gemini CLI did. The relevant benchmark here is Terminal-Bench, which scores an agent driving a real terminal: editing files, running commands, and fixing the failures it hits. Here’s the current top of the Terminal-Bench 2.1 leaderboard.
| Agent | Model | Terminal-Bench 2.1 |
|---|---|---|
| Codex CLI | GPT-5.5 | 83.4% |
| Claude Code | Opus 4.8 | 78.9% |
| Gemini CLI | Gemini 3.1 Pro | 70.7% |
The table shows the specific agent-plus-model pairings discussed here; the full leaderboard ranks more entries. And that pairing matters — a score belongs to an agent plus a model, not the tool alone. OpenCode and Aider are model-agnostic harnesses, so they don’t sit on the leaderboard as a single number — they’re only as strong as the model you connect, and they can run the same Claude or GPT models that top it. So “which is most capable” is really “which agent harness, running which model.” Codex CLI and Claude Code lead today; the open-source tools can get close by pointing at the same flagships.
What actually changed on June 18
Google announced the change at I/O on May 19, 2026, and pulled the plug a month later. The consumer version of Gemini Code Assist on GitHub was deprecated the same day and shut down on July 17.
Check this before you do anything: not everyone is affected. If you have a Code Assist Standard, Code Assist Enterprise, or Code Assist for GitHub subscription managed through Google Cloud, gemini-cli keeps working with no interruption. The shutdown hit the consumer tiers — free, AI Pro, and AI Ultra. If that’s you, read on.
Antigravity CLI — the official path, with a catch
Antigravity CLI is the named successor. It’s a Go-based binary called agy, and migrating is genuinely low-friction.

Install it with the native script:
curl -fsSL https://antigravity.google/cli/install.sh | bash
Then import your old setup. The first time you run agy it detects your Gemini CLI config and offers an interactive migration; you can also do it manually with agy plugin import gemini, which is non-destructive — your ~/.gemini/ directory stays intact, so you can roll back. It pulls over extensions, global config, session tokens, workspace rules, and MCP server definitions. (One gotcha: custom skills in .gemini/skills/ need to be moved to .agents/skills/ by hand.) Google’s migration guide covers the edge cases. If you want to stay inside Google’s tooling, this is the path of least resistance.
The catch is the free tier. Gemini CLI gave free users roughly 1,000 requests per day, reset daily. Antigravity CLI’s free tier comes in around 20 requests per day on a weekly compute cap — developers in Google’s own transition thread reported hitting the wall in a handful of turns, then waiting out a roughly week-long reset. It’s also closed source, where Gemini CLI was Apache 2.0. For paid AI Pro and Ultra users the quotas are more workable, but if the free allowance was the reason you used Gemini CLI, the successor probably won’t replace it.
Antigravity (the IDE) — Google’s flagship agentic platform
The CLI is the drop-in swap, but Antigravity (the IDE) is where Google is actually putting its weight. Google Antigravity is an agent-first IDE — a VS Code fork built around running autonomous agents rather than typing every line yourself. Its Manager surface is a dashboard for spawning and monitoring several agents at once (up to five in parallel), and each agent gets direct access to the editor, terminal, and browser, so it can plan a task, write the code, run it, and check its own work end to end.

It’s not locked to Gemini, either. Antigravity runs Gemini 3 Pro and the faster Gemini 3.5 Flash, but it also supports Anthropic’s Claude models, so you can point the same agent workflow at whichever model you trust for the job.
It’s free during the public preview with generous Gemini 3 Pro quotas, which makes it the more usable free Google option right now if you’re open to leaving the terminal for an editor. Google has signaled a paid tier is coming, so treat the current pricing as a preview perk, not a promise.
OpenCode — the open-source terminal pick
If “free and no subscription” is the whole point, OpenCode is the strongest match for what Gemini CLI’s free tier used to be. It’s an open-source (MIT), terminal-native agent that connects to 75+ model providers and stores everything locally. You bring your own API key, so your only ceiling is your API budget — there’s no per-tool monthly cap to hit.
It’s also the tool with the most momentum in this space: OpenCode crossed 160,000 GitHub stars in early 2026 and the project reports use by millions of developers a month. Model-agnostic by design means you’re not locked to one vendor’s quota decisions — which is exactly the failure mode that just stranded Gemini CLI users.
Install it in about a minute:
curl -fsSL https://opencode.ai/install | bash
It’s also on npm (npm i -g opencode-ai), Homebrew, and Docker.
Claude Code — the best paid terminal agent
If you’ll pay for quality and you want to stay in the terminal, Claude Code is the one to beat. It’s Anthropic’s agentic coding tool: it reads, writes, and runs code across your whole project, not just autocomplete in a file.
There’s no free tier. The entry point is a Claude Pro subscription ($20/month), with Max plans at $100 and $200/month for heavier use, or pay-per-token via the API. Install is one line:
curl -fsSL https://claude.ai/install.sh | bash
(The old npm install -g @anthropic-ai/claude-code route was deprecated in early 2026 in favor of this native installer.) For a direct sense of how it stacks up against an IDE-based assistant, see our Cursor vs Claude Code comparison.
Codex CLI — the other top-tier terminal agent
If Claude Code is one of the two strongest paid terminal agents, Codex CLI is the other. It’s OpenAI’s coding agent, open source under Apache 2.0 and built in Rust, and it does the same job Gemini CLI did: read your codebase, propose multi-file changes, and run commands in a sandbox from the terminal. On capability it’s right at the top: Codex runs GPT-5.5 under the hood, the same model leading Terminal-Bench above, so it’s in the same league as Claude.
On cost it’s more flexible than Claude Code. The CLI itself is free software: you can authenticate with a ChatGPT plan (Codex is included from Plus at $20/month) or bring your own OpenAI API key and pay per token, the same way you’d run Aider. There’s also a limited free trial through the standard ChatGPT free plan — thin for daily use, but enough to test it. Install with the native script or npm:
curl -fsSL https://chatgpt.com/codex/install.sh | sh
# or: npm i -g @openai/codex
If you’re weighing the two flagships, we put them head to head in Claude vs Codex.
Aider — lean and git-native
Aider is the minimalist’s choice. It’s a free, open-source command-line pair programmer that edits files in your local repo and auto-commits each change to git with a generated message. You bring your own API key and pay only for tokens — or run a local model through Ollama for no per-token cost at all.
It assumes you already know git and your stack and want maximum control, which makes it a clean swap for anyone who used Gemini CLI as a scriptable terminal tool rather than a hand-holding assistant. Install it with pip or Homebrew:
pip install aider-chat
# or: brew install aider
We cover it alongside other zero-cost options in our best free AI coding assistants roundup.
Which one should you pick?
- You want the closest thing to the old free quota: OpenCode or Aider. Open source, no subscription, your API budget is the only limit.
- You want to stay with Google and don’t mind an editor: Antigravity (the IDE), free in preview.
- You’re a paid AI Pro/Ultra user who just wants the official swap: Antigravity CLI.
- You’ll pay for the most capable terminal agent: Claude Code or Codex CLI — the two strongest. Pick by ecosystem (Anthropic vs OpenAI/ChatGPT) and the model you trust more.
The broader lesson from the shutdown: a hosted free tier can disappear on a month’s notice. The open-source, bring-your-own-key tools are the ones that can’t be taken away the same way — which is a real point in their favor now, not just a philosophical one.
For more head-to-head breakdowns, see our comparisons of the AI coding tools developers are actually choosing between.
Frequently asked questions
Why was Gemini CLI shut down?
Google retired the consumer tiers. On June 18, 2026, Gemini CLI and the Gemini Code Assist IDE extensions stopped serving requests for free, Google AI Pro, and Google AI Ultra accounts, pointing those users to the new Antigravity CLI. Only paid Code Assist Standard, Enterprise, and Code Assist for GitHub licenses managed through Google Cloud keep working.
Is Antigravity CLI a good replacement for Gemini CLI?
It's the official successor, but the free tier is much smaller. Where Gemini CLI gave free users roughly 1,000 requests a day that reset daily, Antigravity CLI's free tier is capped at around 20 requests a day on a weekly compute window — so if you burn it Monday, you wait until the next Monday. It's also closed source. If you relied on the old free allowance, it's worth trying OpenCode or Aider before settling.
What is the best free Gemini CLI alternative?
For a no-subscription terminal agent, OpenCode and Aider are the strongest free picks. Both are open source, run in your terminal, and let you connect your own API key (or a local model), so there's no monthly cap — you pay only for the tokens you use. OpenCode supports 75+ providers out of the box; Aider is the leaner, git-native option.
Does any alternative keep the same daily free quota Gemini CLI had?
Not on a single hosted free tier. The closest in spirit is going open source with OpenCode or Aider and bringing your own key — your only limit is your API budget — or running a local model through Ollama for zero per-token cost. Antigravity (the IDE) is free during its preview with generous Gemini 3 Pro quotas, which is more generous than the Antigravity CLI free tier.
Which Gemini CLI alternative is the most capable?
On Terminal-Bench 2.1 — the benchmark for agents driving a real terminal — Codex CLI (with GPT-5.5) leads at 83.4%, with Claude Code (Opus 4.8) at 78.9%, both ahead of Gemini CLI's 70.7%. So you're not trading down on ability. OpenCode and Aider are model-agnostic, so they're as capable as the model you connect — point them at the same flagships and they close most of the gap.