← Blog

Gemini CLI vs Claude Code, and Antigravity CLI

October 1, 2026

AI CodingTool Comparisons

Most "Gemini CLI vs Claude Code" articles still rank on one argument: Gemini CLI is free, Claude Code isn't. That argument expired on June 18, 2026. On that date Google stopped serving Gemini CLI requests for free users and for Google AI Pro and Ultra subscribers, and pointed them to a new tool, Antigravity CLI. So the real comparison today has three players, not two:

  • Claude Code — Anthropic's terminal agent. Paid plans or pay-as-you-go API. Closed source.
  • Gemini CLI — still alive, still open source under Apache 2.0, but now only for paid Gemini API keys and organizations on a Gemini Code Assist Standard or Enterprise licence.
  • Antigravity CLI — Google's successor for everyone else, with a usage quota tied to your Google AI plan.

If you're an individual developer who picked Gemini CLI because it cost nothing, you're actually choosing between Claude Code and Antigravity CLI now. If you're on a company licence or a paid API key, Gemini CLI is still a real option. Here's how the three stack up.

Quick Comparison

Claude CodeGemini CLIAntigravity CLI
Who can use itPaid Claude plans, Console API, some cloud providersPaid Gemini API keys, Code Assist Standard/EnterpriseAnyone with a Google account
InstallNative installer, Homebrew, WinGetnpm, npx, Homebrew, MacPortsNative installer script (agy)
ModelsClaude onlyGemini onlyGemini, plus some third-party models by plan
Context window1M tokens on current modelsGemini 3.1 Pro: ~1M input tokensNot published per tool
MCPYesYesYes
Packaged add-onsSkills, pluginsExtensions, skillsPlugins (bundle skills, subagents, MCP, hooks)
Hooks and subagentsYesYesYes
Pricing modelSubscription or metered APIMetered API or org licencePlan-based quota, optional AI credits
Source codeNot open sourceApache 2.0Not open source

What Actually Changed With Gemini CLI

Google announced the change on its developer blog on May 19, 2026. The post, titled an important update on transitioning Gemini CLI to Antigravity CLI, says that on June 18, 2026, Gemini CLI and the Gemini Code Assist IDE extensions stop serving requests for Google AI Pro and Ultra users and for anyone using them free of charge.

Three access routes sit behind that one announcement:

  • Consumer sign-in (free, Pro, Ultra): closed. These users were moved to Antigravity CLI.
  • Organization licence: unchanged. Teams on Gemini Code Assist Standard or Enterprise keep full access.
  • Paid API key: still works. The post says Gemini CLI stays accessible through paid Gemini and Gemini Enterprise Agent Platform API keys.

One trap to watch for. Some of Google's own Gemini CLI pages still carry the old free-tier copy. At the time of writing, the Gemini CLI GitHub repo README still lists a free tier with a personal Google account, even though the Gemini CLI docs site now carries a banner saying Gemini CLI was replaced by Antigravity CLI for unpaid tier and Google One users on June 18th, 2026. Trust the announcement and the banner, not the README's feature bullets.

We covered the wider fallout of this change in our roundup of Claude alternatives, including what it means if you were using Gemini CLI as a free stand-in for Claude Code.

Claude Code: The Paid Terminal Agent With the Deepest Toolkit

Claude Code is Anthropic's coding agent. It reads your codebase, edits files across a project, runs commands, and handles git. The CLI is the full-featured version, and the same engine also runs in VS Code, JetBrains, a desktop app, and the browser.

Install. The Claude Code docs list a native installer (curl -fsSL https://claude.ai/install.sh | bash on macOS, Linux and WSL, plus PowerShell and CMD versions for Windows), Homebrew, and WinGet. Native installs update themselves in the background; Homebrew and WinGet installs don't. If the terminal is the bit that stops you, our Claude Code install guide walks through it.

Models and context. Claude Code runs Claude models, and you can switch between them mid-session. Anthropic's model configuration docs say that on the Anthropic API, Fable 5.1, Sonnet 5 and later, and Opus 4.7 and later run with a 1M-token context window on every plan, including Pro.

Extensibility. This is where Claude Code is strongest:

  • CLAUDE.md project instructions, and it can also read an existing AGENTS.md
  • MCP servers for connecting outside tools and data
  • Skills for repeatable workflows, and plugins for packaging them
  • Hooks that run shell commands before or after actions
  • Subagents and parallel agents coordinated by a lead agent
  • Scheduled routines that run in the cloud

Cloud handoff. claude --cloud starts a new cloud session, and claude --teleport pulls a cloud session into your terminal. Only the desktop app's "Continue in" option sends a local session up to the cloud.

Pricing model. There is no free route. Claude Code is included in every paid Claude plan (Pro, Max, Team, Enterprise) and shares that plan's usage limits, according to Claude's pricing page. The Free plan doesn't include it. You can also pay as you go with a Console API key, which bills by tokens.

Licence. The anthropics/claude-code repo on GitHub is public, but its licence reserves all rights under Anthropic's commercial terms. You can read the issues and changelog; you can't fork the agent.

Gemini CLI: Still Open Source, Now for Paid Access Only

Gemini CLI is an open-source agent that runs Gemini models in your terminal. Nothing about the tool itself was switched off. It still ships regular releases on GitHub (v0.62.0 landed on September 29, 2026), and it's still licensed under Apache 2.0.

Install. npm install -g @google/gemini-cli, npx @google/gemini-cli for a one-off run, brew install gemini-cli, or MacPorts.

Models and context. Gemini models only. Google's API docs list Gemini 3.1 Pro Preview with an input token limit of 1,048,576, so roughly the same 1M ceiling as Claude Code. Context size is no longer a reason to pick one over the other.

Extensibility. More than older comparisons give it credit for. The docs cover MCP servers, extensions, agent skills, hooks, subagents (including remote subagents), a GEMINI.md context file, checkpointing, plan mode and sandboxing. Built-in Google Search grounding is the one feature Claude Code doesn't match out of the box.

Pricing model. Metered. A paid Gemini API key bills per token, or your organization covers it through a Code Assist Standard or Enterprise licence. If you were relying on a personal Google account, that door is closed.

Where it wins. It's the only one of the three you can read, fork, and modify. If your company already pays for Code Assist, or you need an auditable open-source agent, it's still the best Google-side choice.

Antigravity CLI: Google's Successor for Everyone Else

Antigravity CLI is the terminal version of Google Antigravity. Per the Antigravity CLI docs, it installs with curl -fsSL https://antigravity.google/cli/install.sh | bash on macOS and Linux (PowerShell and CMD scripts for Windows) and runs as agy. It shares its agent harness and settings with the Antigravity 2.0 desktop app, and you can export a conversation from the terminal into the app.

Migration. On first run, agy offers a one-time import of your Gemini CLI extensions, skills and settings. Extensions become Antigravity plugins (agy plugin import gemini), skills move from .gemini/skills/ to .agents/skills/, MCP servers move into an mcp_config.json, and the agent still reads your GEMINI.md and AGENTS.md files. The Google blog lists Agent Skills, Hooks, Subagents and Extensions as the features carried over.

Models. Google's models page lists Gemini 3.x models, including Gemini 3.1 Pro, plus a few third-party models such as Claude Sonnet 4.6 and Claude Opus 4.6. Which ones you get, and how much, depends on your Google AI plan. The docs don't publish a context window figure for the CLI itself.

Extensibility. MCP, skills, lifecycle hooks, custom subagents, and plugins that bundle all of those into one package. Built-in subagents include a research agent and a sandboxed browser agent.

Pricing model. According to Antigravity's plans and AI credits page, every account gets a baseline quota. Users without Google AI Pro or Ultra get a weekly-refreshed quota; Pro and Ultra get larger quotas that refresh every five hours, up to a weekly limit. Pro and Ultra users can buy AI credits for overage. The page also says there's currently no bring-your-own-key option for extra rate limits.

Licence. The public GitHub repo holds a README, changelog and examples, not source code, and carries no open-source licence.

Where Each One Is Stronger

If you care most about...Pick
Starting today with no paid planAntigravity CLI
An open-source agent you can audit or forkGemini CLI
Long, multi-file runs and the deepest add-on ecosystemClaude Code
Moving work between terminal, IDE, desktop and webClaude Code
Mixing Gemini with some third-party models in one toolAntigravity CLI
Your company already pays for Gemini Code AssistGemini CLI

Claude Code has the widest reach of the three: one engine across terminal, IDE, desktop, web and cloud routines, with skills, plugins, hooks and subagents all documented in one place. It's the pick when you want to hand off a long, multi-file task and review the result. You pay for that.

Gemini CLI is now a specialist pick: open source, metered, and a natural fit for teams already inside Google Cloud.

Antigravity CLI is the answer to "is there still a free Google terminal agent?" Yes, with a quota. It's also the newest, so expect its docs and behaviour to keep moving.

How to Choose

  1. Check which Gemini CLI route you're actually on. If you sign in with a personal Google account, you're an Antigravity CLI user now.
  2. Decide whether you'll pay. If not, Antigravity CLI is the only one of the three with a no-subscription route.
  3. Decide whether open source matters. Only Gemini CLI qualifies.
  4. Run the same real task through two of them. Pick a multi-file change you did last month. How each agent handles a long run tells you more than any feature table.
  5. Don't choose on context size. All three now sit at or near 1M tokens on their flagship models.

If you're also weighing OpenAI's agent, our Claude Code vs Codex comparison covers that matchup.

Where Taku Fits

All three tools assume you're comfortable in a terminal and willing to set things up. If what you actually want is to use something a power user already built with them, like a skill, an agent workflow or a small AI app, Taku is an AI-native desktop workspace where you find it, mirror it into your own workspace, run it, and remix it, without setting up GitHub repos, dependencies or API keys yourself. It isn't a coding agent and doesn't replace one. Taku is in Beta, and the Mac app is available now; there's also an experimental, unsigned Windows build for Windows 10 and 11.

FAQ

Is Gemini CLI still free?

No. Since June 18, 2026, Gemini CLI no longer serves free users or Google AI Pro and Ultra subscribers. Access continues through paid Gemini or Gemini Enterprise Agent Platform API keys, and for organizations on a Gemini Code Assist Standard or Enterprise licence. The free Google option is now Antigravity CLI, which has a weekly-refreshed baseline quota for accounts without a paid Google AI plan.

What is the difference between Antigravity CLI and Claude Code?

Antigravity CLI runs mainly on Gemini models, uses a quota tied to your Google AI plan, and shares its agent with the Antigravity 2.0 desktop app. Claude Code runs Claude models only, needs a paid Claude plan or API key, and works across terminal, IDE, desktop and web. Both support MCP, skills, hooks and subagents.

What is the best Gemini CLI alternative?

For most individuals, Antigravity CLI, because it's Google's official replacement and imports your Gemini CLI extensions, skills and settings. If you're willing to pay and want the most mature terminal agent, Claude Code. If you specifically need open source, stay on Gemini CLI with a paid API key.

Is Claude Code open source?

No. The anthropics/claude-code GitHub repo is public for issues and the changelog, but the licence reserves all rights under Anthropic's commercial terms. Gemini CLI is the open-source option here, under Apache 2.0.

Which terminal AI coding agent has the biggest context window?

They're effectively tied. Claude Code's current models run with a 1M-token window on the Anthropic API, and Gemini 3.1 Pro has an input limit of 1,048,576 tokens. Antigravity doesn't publish a separate figure for its CLI. How well an agent uses its context on a long task matters more than the ceiling.