by Google

Antigravity CLI pricing, plans and limits

Google's terminal agent for multi-file coding, subagents and browser checks.

Best for
Backend and full-stack engineers making scoped changes to an existing repository from the terminal
Not for
Teams needing SOC 2 or GitLab/Bitbucket support
  • Coding
  • Research

Should you use Antigravity CLI?

Built for developers who want Gemini CLI's terminal workflow with more model choice and background subagents. Its picker spans five models, Gemini and Claude and GPT-OSS-120B, and it suits scoped edits to an existing codebase better than building an app from a blank folder.

A model picker that spans Gemini, Claude and an open-weight model in one terminal session, so a team is not locked into a single vendor's models.

Deep integration with Google's model lineup and free Google-account access gives it distribution few rivals can match, but the free-tier quota cuts have already pushed some users toward Claude Code and Codex CLI.

Also for

  • Engineering teams who want one agent writing code while subagents run builds or research in parallel
  • Developers who want to switch between Gemini and Claude models without changing tools
Show 3 more
  • Former Gemini CLI users migrating after its June 2026 consumer shutdown
  • Full-stack engineers doing scoped edits in an existing repo
  • Teams already using Gemini or Google Cloud tooling

Also not for

  • Developers building an app from a blank folder

Main trade-off

Free-tier quota (roughly 20 requests/day) is a fraction of what Gemini CLI offered, and Google has publicly acknowledged provisioning has not kept up with demand since April 2026.

Antigravity CLI pricing

Top plan
$199.99/mo
Billing model
Freemium

Free for individuals with a Google account (small daily agent-request quota). Pro is $19.99/month, Ultra 5x is $99.99/month, and Ultra 20x is $199.99/month for higher usage caps. AI credit overages cost $0.01 each, sold in bulk at $199 for 20,000 credits.

Before you commit

  • AI credit overages at $0.01 each once a plan's quota is exhausted
  • No SOC 2 report, which can block enterprise procurement

What Antigravity CLI does on its own

Scope of autonomy
Semi-autonomous
Human approval boundary
Human in the loop
Maturity
Generally available

Safety controls

  • Runs sandboxedSupported
  • Human in the loopSupported
  • Audit loggingNot supported
  • Output filteringNot supported

What it can operate

  • Web browsing
  • Code execution
  • File system
  • Shell commands
  • Document reading

Where Antigravity CLI runs

Deployment
Local · Cloud · API
Where it runs
CLI · API · IDE plugin
Memory
Long term
Context window
1M tokens

Models it runs on

  • Gemini 3.1 Pro
  • Gemini 3.8 Flash
  • Claude Sonnet 4.6
  • Claude Opus 4.6
  • GPT-OSS-120B

Systems it connects to

SystemType
GitHubNative
Google ChromeNative
Model Context ProtocolNative
JetBrains IDEsPlugin

Languages it supports

  • Python
  • JavaScript
  • TypeScript
  • Go
  • Java
  • Rust

Benchmark scores

Terminal-bench 2.1 (Gemini 3.8 Flash, agentic terminal coding)
89.4 / 100
DeepSWE v1.1 (Gemini 3.8 Flash, long-horizon software engineering)
73.7 / 100
Terminal-bench 4.0 (Gemini 3.8 Flash, general agent capability)
19.1 / 100

What Antigravity CLI does

Antigravity CLI is Google's terminal AI agent, released as the CLI surface of the Antigravity 2.0 platform on May 19, 2026, replacing Gemini CLI. It runs coding tasks through a multi-agent system with independent subagents, and unlike its Gemini-only predecessor its model picker also adds Claude and an open-weight model.

  • Subagents run background work in parallel

    The main agent spawns independent subagents for tasks like running a build or looking up documentation, each starting with a clean context and either inheriting the workspace, branching into an isolated Git worktree, or sharing storage.

  • Five-model picker, not Gemini-only

    Free, Pro and Ultra plans can switch between five models spanning two vendors plus an open-weight option from the same session, a wider lineup than Gemini CLI ever offered.

  • Browser-verified UI changes

    A controlled Chrome instance lets the agent edit a UI, open the page itself, and confirm the change actually rendered before reporting the task done.

  • Plan-before-you-build mode

    Typing /plan makes the agent draft an editable plan for review before it writes any code, with a Plan Review Policy setting controlling when a plan needs sign-off.

Show all 6 features
  • Three-tier permission model

    request-review pauses for approval on every command by default, proceed-in-sandbox contains actions in a virtualized container, and always-proceed runs on the host machine after one first-time approval per tool class.

  • Live MCP manager

    The /mcp command opens an interactive manager that configures, monitors and shows live connection status for Model Context Protocol servers instead of requiring a manual config-file edit.

Strengths and limits

Strengths

  • It supports models from three different vendors instead of Gemini alone, so a team is not locked to one vendor's models the way Gemini CLI was.
  • Subagents can branch into an isolated Git worktree, so a background build or lookup task cannot corrupt the main session's working directory.
  • Individual access remains free with a Google account, so a team can evaluate it before deciding whether to pay for higher quota.

Limits

  • The free tier's daily agent-request quota is far below the roughly 1,000 requests a day Gemini CLI granted free users before its June 2026 shutdown, and Google has acknowledged since April 2026 that demand exceeds its free-tier provisioning.
  • Google has not published a SOC 2 report for Antigravity, and native repository support is GitHub only, with no GitLab or Bitbucket integration.
  • Reviewers report it is weakest at building an app from an empty folder; scoped edits to an existing codebase perform noticeably better than greenfield builds even with detailed prompts.

Antigravity CLI alternatives

AgentWhere it runsPriceMaturityChecked
Antigravity CLI This pageCloud · Your machine · APIFree planAvailable27 Sep
Gemini CLIPick Antigravity CLI for the actively maintained, still-free successor; Gemini CLI itself no longer serves free or Pro/Ultra requests.Your machine · APIUsage-basedEnterprise27 Sep
Claude CodePick Claude Code if you want the most thorough multi-file PR workflow; pick Antigravity CLI if you want to switch between Gemini and Claude models in the same session.--------
OpenAI CodexPick OpenAI Codex if you already pay for ChatGPT and want the tightest CI/CD integration; pick Antigravity CLI if you want a free entry point and a broader model picker.Cloud · Your machine · APIFree planEnterprise6 Nov

Also compared

  • JulesPick Jules if you want an async agent that works on a ticket while you do something else; pick Antigravity CLI if you want an interactive terminal session with live subagent status.

Antigravity CLI facts

Works offline
No
Minimum skill level
Intermediate
Budget fit
Free
Open source
Yes
Ships
Weekly
Last major update
Maker founded
2025
Headquarters
Mountain View, CA

What changed

  • New feature v1.2.11: selectable reasoning effort, plugin and rendering fixes source
  • Release Antigravity 2.0 and Antigravity CLI launch at I/O 2026 source

Questions about Antigravity CLI

How much does Antigravity CLI cost in 2026?

Antigravity CLI itself costs nothing to start, but Google's paid tiers unlock higher usage: Pro is $19.99 a month, Ultra 5x is $99.99 a month, and Ultra 20x is $199.99 a month. Heavy users who exceed their plan's quota can add AI credits at a cent apiece, with a 20,000-credit bulk pack priced at $199.

Does Antigravity CLI have a free plan?

Yes: the free tier grants roughly 20 agent requests a day as of September 2026, refreshing every 5 hours once exhausted, a steep drop from the roughly 1,000 a day Gemini CLI gave free users before its June 2026 shutdown. Google has publicly acknowledged since April 2026 that demand for the free tier has outpaced its provisioning, so refresh delays are common.

Which tools compete with Antigravity CLI in 2026?

Claude Code produces more thorough multi-file pull requests but requires its own paid subscription with no comparable free tier. OpenAI Codex CLI has the tightest CI/CD integration and is effectively free for existing ChatGPT subscribers. Jules, also from Google, works asynchronously on a ticket in the background rather than in an interactive terminal session.

What separates Antigravity CLI from Gemini CLI?

Antigravity CLI is Gemini CLI's direct successor: Gemini CLI stopped serving free, Pro and Ultra requests on June 18, 2026, while Antigravity CLI keeps a free tier open to individuals. It also adds a multi-agent subagent system and a model picker that includes Claude Sonnet 4.6, Claude Opus 4.6 and GPT-OSS-120B, none of which Gemini CLI ever offered.

How do you get started with Antigravity CLI?

Install it from the google-antigravity/antigravity-cli GitHub repository or the platform's own installer, then sign in with a Google account to activate the free tier. Run /plan on your first task to have the agent draft an editable plan before touching any files, and use /mcp to connect any Model Context Protocol servers your workflow needs.

About Antigravity CLI

Antigravity CLI is Google's terminal agent for the Antigravity 2.0 platform, the agent-first development environment Google unveiled in November 2025 and expanded at I/O 2026 on May 19, 2026. It replaced Gemini CLI as Google's primary terminal coding agent after Gemini CLI stopped serving free, Google AI Pro and Google AI Ultra requests on June 18, 2026. Where Gemini CLI ran a single agent through one conversation, Antigravity CLI runs on the same multi-agent system as the Antigravity desktop app, spawning independent subagents that can inherit the current workspace, branch into an isolated Git worktree, or share storage while the main session keeps working.

Read the full background

The CLI defaults to Gemini 3.1 Pro and Gemini 3.8 Flash, but its model picker also lists Claude Sonnet 4.6, Claude Opus 4.6 and the open-weight GPT-OSS-120B on Free, Pro and Ultra plans, a wider lineup than the Gemini-only rival it replaced. Three permission levels set how much it can do alone: request-review pauses before every command by default, proceed-in-sandbox lets it act inside a virtualized container, and always-proceed hands it the host machine after a single first-time approval per tool class. It connects to Model Context Protocol servers through an interactive /mcp manager, and its Chrome integration lets it change a UI, open the browser itself, and confirm the change actually rendered.

It suits developers already working inside an existing codebase who want scoped, incremental changes reviewed and executed from the terminal, plus teams who want one agent writing code while background subagents run builds or look up documentation in parallel. It is weaker at building an app from a blank folder: reviewers testing greenfield builds with detailed prompts report noticeably worse results than its scoped-edit performance. Reviewers most often set it against Claude Code, OpenAI Codex and Google's own async agent, Jules.

Individual access is free with a Google account, with Pro, Ultra 5x and Ultra 20x plans unlocking higher usage caps, and AI credit overages available for heavy users. It runs on macOS, Linux and Windows (including WSL), ships as a public GitHub repository at google-antigravity/antigravity-cli, and reuses the same ~/.gemini configuration directory Gemini CLI used.

Google shipped v1.2.11 on September 25, 2026, adding a selectable --effort reasoning level and fixing Mermaid and LaTeX rendering in WezTerm and VS Code's integrated terminal; recent weeks also added a /plan mode that drafts an editable plan before writing code and a /boost command that runs a multi-agent reasoning pipeline on hard problems. Google has publicly acknowledged quota strain since April 2026, as demand outpaced its free-tier infrastructure.

Sources and guides

More AI Agents on Hokai

  • OpenHandsOpen-source AI agent that plans, codes, tests, and ships pull requests
  • PromptlessAI teammate that drafts doc updates as pull requests when you ship.
  • QualGentAutonomous QA agent that tests iOS and Android apps like a human.
  • OpenCodeOpen source coding agent that works with any model, in terminal, desktop or IDE.
  • Codex SecurityOpenAI's security agent that scans repos and drafts fixes.
  • ClineOpen source AI coding agent that shows every edit before it runs.