Claude Code vs Codex CLI vs Antigravity CLI: Limits

Three terminal coding agents, Claude Code, Codex CLI and Antigravity CLI, compared on plans, usage limits and sandbox controls

Terminal coding agents now come bundled into subscriptions, and each vendor meters them differently. This post compares Anthropic's Claude Code, OpenAI's Codex CLI and Google's Antigravity CLI on what plan unlocks them, how usage is counted, and how each one contains a command it runs on your machine. It also explains why Gemini CLI, the tool many comparisons still list, is no longer a fair third entry.

This comparison is based on official documentation, published specifications, and vendor pricing verified as of November 1, 2026. It is not based on hands-on use. OpenAI's Codex documentation could not be opened from where this was researched, so Codex rows are thinner and marked accordingly.

Why Antigravity CLI replaces Gemini CLI here

Google's transition post says that on June 18, 2026, Gemini CLI stopped serving requests for Google AI Pro and Ultra users and for free users of Gemini Code Assist for individuals. Organizations with a Code Assist Standard or Enterprise license keep it, and it remains reachable through paid API keys. Antigravity CLI is the successor, built in Go on the same agent harness as the Antigravity 2.0 desktop app, and Google states there will not be "1:1 feature parity right out of the gate."

One official page still contradicts this. The Gemini CLI quota page lists 1,000 requests a day for personal Google accounts, 1,500 for Google AI Pro and 2,000 for Ultra, while the same page says the CLI was replaced on June 18. Those numbers describe a service that has ended for individuals, so they are not used in the tables below.

What unlocks each tool, and what it costs

Item	Claude Code	Codex CLI	Antigravity CLI
Free access	No. Claude's pricing page lists Claude Code as not included in Free	Limited. The ChatGPT plans page marks Codex "limited" on Free	Yes. An account with no Pro or Ultra plan gets a weekly quota, per the Antigravity plans page
Cheapest full plan	Pro: $20 monthly or $17 a month billed annually	ChatGPT Plus: $20, with "expanded Codex usage"	Google AI Pro: $19.99
Heavy tier	Max, from $100 (5x or 20x Pro)	Pro, $100 (5x Plus Codex usage) and $200 (20x, relayed)	Google AI Ultra: $99.99 (5x Pro) and $199.99 (20x Pro)
How limits reset	Rolling five-hour session plus a weekly limit, shared with Claude chat	Not stated on the pages read	Five-hour refresh on Pro and Ultra until a weekly limit; weekly refresh with no plan
Overage	Usage credits at API rates	Not stated on the pages read	AI credits on Pro and Ultra only

The Google prices come from the Gemini subscriptions page, and the Pro tier launch is described in an OpenAI Developer Community announcement. That thread does not state the $200 price itself, so treat it as relayed. Plus's $20 per month comes from OpenAI's help center, which names no country, and the ChatGPT plans page itself shows no dollar amounts.

Normalizing the heavy tiers

All three vendors sell a "5x" and a "20x" step, but against different bases. Claude's Max multiplies Pro, ChatGPT Pro multiplies Plus Codex usage, and Google's Ultra multiplies Pro. Anchoring each on its own roughly $20 plan gives these derived estimates, which measure vendor-defined usage and not tasks or tokens:

Tier	Price	Multiple of the ~$20 plan	Derived price per ~$20-plan unit
Claude Max 5x	from $100	5x	about $20
Codex via ChatGPT Pro $100	$100	5x Plus Codex	about $20
Codex via ChatGPT Pro $200 (relayed)	$200	20x Plus Codex	$10 per unit
Antigravity via Ultra 5x	$99.99	5x Pro	about $20
Antigravity via Ultra 20x	$199.99	20x Pro	about $10

The pattern is the same everywhere: the first step up costs the same per unit as the base plan, and the 20x step halves the unit price. Claude's 20x price is not shown on its pricing page, so that row is left out. The structure is a vendor-wide convention, but nothing here shows the raw quotas are equal. None of the three publishes request or token counts for these plans.

A second caveat applies to Google. The same Ultra subscription also covers the Gemini app, 20 TB or more of storage and other products, so the whole $99.99 is not a coding-agent price. A Claude Max or ChatGPT Pro subscription is shared with chat in the same way. Claude's support page says chat and Claude Code draw on the same pool.

How each one contains a command

Control	Claude Code	Codex CLI	Antigravity CLI
Default prompting	Normal permission flow for each Bash command	Not verified: Codex docs unreadable	Default mode asks before file writes and terminal commands
OS sandbox	Off by default; enable with `/sandbox`	Sandbox plus separate approval policy; OpenAI describes read-only and workspace-write modes	Off until enabled by `--sandbox` or auto mode
Network	No direct route; hosts must be added to an allowlist that starts empty	Not open-ended in OpenAI's internal setup; defaults for users not verified	Off for CLI commands by default
Windows	Native Windows unsupported for the sandbox; use WSL2	Not verified	Needs Windows 11 24H2 with the August 2026 update
Auto mode	Auto-allow for sandboxed commands	Auto-review via an approval subagent (OpenAI's description)	Auto (Preview), not available to business or enterprise accounts

The Claude details are from its sandboxing documentation, the Antigravity details from its execution modes and terminal sandbox pages, and the Codex row from OpenAI's post Running Codex safely at OpenAI. The Codex post describes OpenAI's own deployment, so it shows the design, not the defaults a user sees.

Two traps stand out. Claude Code's own documentation warns that without failIfUnavailable, it silently runs commands unsandboxed when the sandbox cannot start, which is the situation on native Windows. Its built-in Read, Edit and WebFetch tools also run outside the sandbox and follow permission rules instead. In Antigravity, a headless run with --mode=auto prints a warning and falls back to default mode.

Setup differences that cost time

Claude Code installs with a native script, Homebrew, WinGet or Linux package managers. If an ANTHROPIC_API_KEY environment variable is set, it is used instead of your subscription and API charges apply, which Claude's support page calls the most common cause of unexpected charges. Max's monthly API credits cannot be spent in Claude Code or the Claude apps.
Antigravity CLI keeps settings in ~/.gemini/antigravity-cli/settings.json. The plans page says bring-your-own-key and bring-your-own-endpoint are not supported, and third-party models come only with Ultra.
Codex CLI is one of three ways to use Codex, alongside the ChatGPT app and an IDE extension, all connected through a ChatGPT account, per OpenAI's Codex page. That page gives no install method, model or limit.

Who each one does not fit

Claude Code: anyone who wants a free tier, anyone on native Windows who needs a real sandbox, and anyone expecting Max's bundled API credits to pay for it, since those cannot be spent in Claude Code.
Codex CLI: readers who need published quotas or documented defaults, since neither could be verified here. Its Free access is explicitly limited.
Antigravity CLI: teams that need feature parity with Gemini CLI today, business or enterprise users who want Auto mode, and anyone who needs exact quota numbers, because the plans page publishes none.
Gemini CLI itself: personal-account users after June 18, 2026.

Which one for which situation

If you want	Look at	Why, from the published pages
To try a terminal agent for $0	Antigravity CLI or Codex CLI	Both document free access; Claude Code does not
A documented OS sandbox with a hostname allowlist	Claude Code	Its sandbox page spells out filesystem, network and credential handling in detail
Third-party models in the same agent	Antigravity CLI on Ultra	The plans page lists them as Ultra-only
One bill that also covers chat and storage	Any of the three subscriptions	Each shares its pool with the vendor's chat app, so check overlap before buying
A tool that works in CI and scripts	Claude Code	Its docs show `claude -p` piping and GitHub Actions and GitLab CI/CD guides; the other two were not verified
Enterprise policy lock-in	Claude Code or Codex	Both describe admin-enforced settings; Antigravity's Auto mode is unavailable to business accounts

FAQ

Is Gemini CLI still available?

Not for personal accounts. Since June 18, 2026 it no longer serves Google AI Pro, Ultra or free Code Assist for individuals users, though Standard and Enterprise licenses and paid API keys keep access.

Can I use Claude Code on the free plan?

No. Claude's pricing page lists Claude Code as available from Pro, at $20 a month or $17 a month billed annually.

Which tool has the largest published usage limit?

None can be ranked. Claude, ChatGPT and Google each quote multiples of a base plan and none publishes request or token counts for these plans.

Last verified: November 1, 2026. Quotas and plan structure change often, and the Google CLI lineup was mid-transition. Pages that could not be read: OpenAI's Codex documentation under developers.openai.com/codex (redirected to a site that was unreachable), OpenAI's help article on using Codex with a ChatGPT plan (fetch failed) and the openai/codex repository on GitHub (unreachable).