Gemini CLI vs Claude Code: Which to Pick

If you are choosing today, the first fact to know is that Google has moved its individual users off Gemini CLI. The README says Gemini CLI is now for organizations with Gemini Code Assist Standard or Enterprise on Google Cloud. For everyone else, Google points at Antigravity CLI. Claude Code is the more complete product on a Claude plan. Facts below were checked on 31 October 2026.

The short version

What changed for Gemini CLI

An older comparison would tell you that Gemini CLI is the free terminal agent with a generous daily allowance. That is out of date. The top of the project's README now carries a notice with four points:

• It is designed for Gemini Code Assist Standard or Enterprise licenses and Code Assist for GitHub on Google Cloud.
• Free individual access and Google AI Pro and Ultra access moved to Antigravity CLI on June 18, 2026.
• It is not tuned for the Gemini 4 series or later.
• Do not use it to benchmark Gemini models, because the harness is not tuned for Gemini 4 and later and produces unrepresentative scores.

The README includes a small table of "which Google CLI do you need." Antigravity CLI (agy) is the interactive multi-agent coding terminal for Gemini 4 Argon and later, free accounts, Google AI Pro and Ultra accounts, and Google Cloud. A third tool, the Gemini API CLI (gemini-api), is a stateless API command line that is explicitly not a coding harness. Gemini CLI itself is listed for organizations with Code Assist licenses, plus legacy paid API and Vertex keys.

For a reader comparing it with Claude Code, that means the real question is narrower than the title suggests. Do you work somewhere that already pays Google for Code Assist? If yes, Gemini CLI is a live option. If no, you are probably choosing between Claude Code and Antigravity CLI.

Side-by-side facts

Installing and signing in

Gemini CLI

Run it once without installing:

npx @google/gemini-cli

or install it globally:

npm install -g @google/gemini-cli
# or
brew install gemini-cli

The README lists three ways to authenticate.

1. Sign in with Google. For a paid Code Assist license from your organization, set your Google Cloud project first: export GOOGLE_CLOUD_PROJECT="YOUR_PROJECT_ID", then run gemini.
2. Gemini API key. Get a key from Google AI Studio and export GEMINI_API_KEY. The README pitches this for developers who need specific model control or paid-tier access, with usage-based billing.
3. Vertex AI. Export GOOGLE_API_KEY and GOOGLE_GENAI_USE_VERTEXAI=true. This is the enterprise route, with higher rate limits on a billing account.

Pick a model with a flag, for example gemini -m gemini-2.5-flash. Add extra directories with --include-directories ../lib,../docs.

Claude Code

curl -fsSL https://claude.ai/install.sh | bash
claude --version
claude

On first use it prompts you to log in. If ANTHROPIC_API_KEY is set and you approve it, the prompt is skipped. Native installs update in the background. Homebrew and WinGet installs do not, so you run brew upgrade claude-code or winget upgrade Anthropic.ClaudeCode yourself.

Cost

This is where the two diverge most.

Gemini CLI. There is no consumer plan to buy through it any more. Your cost is whatever your organization pays for Code Assist, or the per-token charge on a Gemini API key or Vertex AI project. We did not find a published per-seat price for Code Assist Standard or Enterprise in the README, so we do not quote one.

Claude Code. Included with Claude Pro, Max and Ultra:

Claude Code is not on the Free plan. The pay-per-token route is available through an Anthropic Console account. Anthropic's cost page says that across enterprise deployments the average is around $13 per developer per active day, or $150 to $250 per developer per month, with 90 percent of users under $30 per active day.

Features that matter in daily use

Project memory and instructions

Both read a markdown file at the root of the repo. Gemini CLI uses GEMINI.md. Claude Code uses CLAUDE.md, and if your repo already has an AGENTS.md for other agents it can read that instead. If you work across several agents, AGENTS.md is the least repetitive option, and Claude Code's support for it is documented.

Extensions, plugins and skills

Gemini CLI has custom commands and extensions that you build and share. Antigravity's migration doc notes that the industry has settled on the word "plugins," and converts Gemini CLI extensions to plugins and some commands to skills on import.

Claude Code has skills (a SKILL.md file that adds a repeatable workflow), subagents (each in its own context window with its own tool access), hooks (shell commands, HTTP endpoints, MCP calls, prompts or subagents fired at lifecycle events) and plugins. The hooks run the same wherever Claude Code runs: terminal, IDE extensions, desktop app and cloud sessions.

Safety controls

Gemini CLI documents trusted folders (execution policies per folder), a sandbox, and checkpointing to save and resume conversations. Claude Code asks before edits and non-trivial shell commands, has a read-only plan mode, matches bash permission rules against the whole command (so an allow rule for one command does not approve a chained second one), and has a sandbox with managed settings for organizations.

Scripting

Both have a non-interactive mode. Gemini CLI: gemini -p "Explain the architecture of this codebase" --output-format json, and --output-format stream-json for newline-delimited events on long jobs. Claude Code: claude -p "...", for example git diff main --name-only | claude -p "review these changed files for security issues". Both are fine for CI.

Models

Gemini CLI is for Gemini models. The README tells you the harness is not tuned for Gemini 4 and later, which is a notable thing for a vendor's own CLI to say. If the newest Gemini model matters to you, the README's own answer is Antigravity CLI.

Claude Code runs Claude models. The model docs describe aliases (opus, sonnet, haiku, fable, best) and an opusplan mode that uses Opus in plan mode and Sonnet for execution. The Terminal CLI, VS Code and JetBrains also support third-party providers, so a company that buys Claude through Amazon Bedrock or Google Vertex can still use the same agent.

Data handling

For Claude Code, Anthropic's data-usage page says that Free, Pro, Max and Ultra accounts have a setting that lets data be used to improve future models, and that it covers Claude Code use on those accounts. The page gives a five-year retention period for users who allow it. For Team, Enterprise, API and third-party platform use, Anthropic says it does not train generative models on Claude Code code or prompts under commercial terms unless the customer opts into a program.

For Gemini CLI, the policy depends on how you sign in, because Google's terms differ between Code Assist, the Gemini API and Vertex AI. We did not read each of those terms for this page.

Moving off Gemini CLI

If you are a Gemini CLI user whose access ended, Google documents a migration path. The first time you run agy in an environment that has Gemini CLI configuration, it detects it and offers a checklist. You choose which assets to convert; session tokens move into the operating system's keyring; settings map across. Skills, rules and MCP servers are kept. The doc warns of partial parity, since some custom terminal themes or experimental overlays may not carry over. Context files and workspace rules use identical rules on both CLIs, so existing rule documents need no change.

If you would rather change vendors than change Google CLIs, the move to Claude Code is mostly manual:

1. Rename or copy GEMINI.md to CLAUDE.md (or keep a shared AGENTS.md).
2. Re-add each MCP server with claude mcp add.
3. Recreate custom commands as skills. Claude Code merged custom commands into skills: a file at .claude/commands/deploy.md and a skill at .claude/skills/deploy/SKILL.md both create /deploy.
4. Re-set permissions in Claude Code's settings.

Where Claude Code is stronger

• More surfaces from the same engine: terminal, VS Code, JetBrains, desktop app, web, and a phone via Remote Control.
• Scheduled work. Cloud scheduled tasks keep running when your computer is off.
• A bigger, documented extension model: skills, subagents, hooks, plugins, and an Agent SDK for building custom agents.
• A clear consumer path. Anyone can buy a plan and get the agent.

Where Gemini CLI is stronger

• It is open source under Apache-2.0, so you can read, audit and fork it, and the README invites contributions.
• You pick the backend. A Gemini API key gives model control, Vertex AI gives enterprise controls, and a Code Assist license fits a Google Cloud estate.
• It fits Google Cloud shops that want one billing and admin path.
• It runs through npx with no install step, which suits locked-down machines.

Which to pick

Pick Gemini CLI when:

• Your company holds Gemini Code Assist Standard or Enterprise licenses.
• You must be able to audit or modify the agent's code.
• You already deploy on Vertex AI and want everything on one cloud contract.

Pick Claude Code when:

• You are an individual or small team without a Google Cloud agreement.
• You want one agent across terminal, IDE, desktop and web.
• You already hold a Claude Pro or Max plan.
• You rely on hooks, skills and subagents to automate your process.

Look at Antigravity CLI when:

• You liked Gemini CLI as a free or Google AI plan tool. That is the product Google now points those users to.

A short checklist before you commit

• Check which Google CLI your Google plan actually covers. The README table is the reference.
• Confirm your organization's data terms for whichever sign-in method you use.
• Make a shared AGENTS.md if you plan to try more than one agent.
• Try one real task on each, with the same prompt, and read the diffs. We did not run that test for this page.

Our take

The documented facts say these are no longer two equal consumer tools. Google has split its offering: Gemini CLI for Code Assist customers, Antigravity CLI for most everyone else. That makes the honest comparison for an individual "Claude Code or Antigravity," and the honest comparison for a Google Cloud enterprise "Gemini CLI or Claude Code through Vertex AI."

On features, Claude Code's documentation is broader: more surfaces, a scheduling story, and an extension system built from skills, hooks and subagents. On openness, Gemini CLI wins clearly, since Claude Code's repository is all rights reserved.

We have not run the two on the same repository, so we make no claim about code quality or speed. If you decide on openness or on your cloud contract, the answer is already in the tables above. If you decide on workflow features, read the Claude Code docs for skills and hooks and compare them with what Gemini CLI's custom commands and extensions offer.

Related

• Antigravity vs Claude Code
• Codex vs Gemini CLI
• Gemini CLI
• Claude Code
• Open source AI agents
• How we compare