Gemini vs Claude

Short answer: for most people asking this, the model is not the part that changed — the tooling is. Google stopped serving Gemini CLI to individual users on 18 June 2026 and replaced it with Antigravity. If you tried Gemini for coding before that date and gave up because the tool annoyed you, you formed an opinion about a product that is no longer the one on offer.

So this page does not referee a benchmark. It covers what you actually drive each model with, what each company does and does not publish about price, and which of those facts is stable enough to decide on.

What the question usually means

"Gemini vs Claude" is two questions wearing one coat. The first is about models: which writes better code. That has a real answer on any given week, and it changes on a schedule you do not control.

The second is about what you drive the model with — the terminal agent, the editor, the subscription it is attached to. That is the part you live in daily, and it decides whether the model's ability ever reaches your repository. It is also the part almost nobody compares, and it is where this year's real change happened, on Google's side.

What happened in June

On 19 May 2026 Google announced that Gemini CLI was being transitioned to Antigravity CLI. The consumer cut-off landed a month later:

On June 18, 2026, Gemini CLI and Gemini Code Assist IDE extensions will stop serving requests for Google AI Pro and Ultra, as well as those using it free of charge using Gemini Code Assist for individuals.

Gemini Code Assist for GitHub went the same way: no new installations from that date, with requests ceasing "in the following weeks." Enterprise is the exception, and a clean one:

If your organization uses Gemini CLI or our IDE extensions via a Gemini Code Assist Standard or Enterprise license, or if your organization uses Gemini Code Assist for GitHub through Google Cloud, your access remains unchanged.

Google added that Gemini CLI "will remain accessible via paid Gemini and Gemini Enterprise Agent Platform API keys."

One more detail matters if you valued the old tool's openness. The repository google-gemini/gemini-cli is Apache-2.0, is not archived, and was still receiving pushes on 2026-07-31 — consistent with continued enterprise support. The successor repository, google-antigravity/antigravity-cli, held a README, a changelog, a demo GIF and an examples directory when we checked the same day: no source tree, no licence. Google's announcement calls the new CLI "Built in Go" but says nothing about licensing. So the replacement is something you download and run rather than read or fork — the most important line on this page, if that is why you picked the old one.

What Antigravity actually is

Antigravity is not a CLI with a new name but a four-surface suite in which the CLI is one part. Versions from Google's documentation on 2026-07-31:

The pieces share a backend with the desktop app, which is the stated reason for the migration: improvements to the core agents land everywhere at once. Google says Agent Skills, Hooks, Subagents and Extensions carried over — Extensions "now as Antigravity plugins" — while conceding there would not be immediate one-to-one feature parity at launch.

Those names will look familiar if you have used Claude Code, because they are broadly the same primitives under the same words. That is convergence rather than coincidence: deterministic hooks and subagent fan-out are becoming table stakes generally.

How Claude's side is shaped

Anthropic's side is narrower and, right now, more stable. Claude Code is a terminal agent with an IDE integration: one of it, from one vendor, talking to one company's models, bundled with a Claude subscription rather than sold as its own SKU. It has not been renamed or replaced this year.

That narrowness cuts both ways. It means fewer surfaces to choose between and a short path from "I pay for Claude" to "I have an agent" — the Pro plan's feature list simply says it includes Claude Code. It also means you pick a vendor and a tool in one move: Claude Code is how you use Anthropic models, not a neutral harness. Antigravity's multi-model roster is a real difference in shape here, in Google's favour.

If you want a deliberately vendor-neutral harness, neither is quite it — that belongs to the editor-versus-agent question instead.

What each side publishes about price

The two differ most concretely here, and in disclosure as much as amount.

Anthropic publishes exact figures. From the Claude pricing page on 2026-07-31: Free at $0; Pro at $17/month billed annually, $20 monthly; Max "From $100" per month, offered as 5× and 20× usage tiers; Team at $20 per seat/month billed annually ($25 monthly), with a premium seat at $100 annually ($125 monthly); Enterprise at $20/seat plus usage at API rates. Claude Code is included rather than added on. The page carries its own caveat that plans may change.

Google publishes structure but not most of the numbers. The Antigravity pricing page shows four tiers — For Individuals, Google AI Pro, Google AI Ultra, and a new Organization plan via Google Cloud. Only one carries a figure: For Individuals is listed at "$0/month", described as "Experience Antigravity without a subscription plan", with unlimited Tab completions, unlimited Command requests and "Basic weekly rate limits". Pro and Ultra carry no price — they route to a Google AI subscription priced per country, so there is no single figure to quote. The Organization plan is described as "Consumption-Based API Pricing".

The documentation tells the same story. Antigravity's plans page states no dollar amounts anywhere, describing quotas qualitatively — Ultra gets "The highest, most generous quota, refreshed every five hours", Pro a "High, generous quota", everyone else a "Meaningful quota, refreshed weekly". Overage runs on purchased AI credits at Gemini Enterprise Agent Platform consumption rates, which that page does not reproduce either. It also lists bring-your-own-key and bring-your-own-endpoint as not supported.

So we are not going to turn this into a monthly cost comparison: from published sources, doing so honestly is not possible. Anyone showing you a clean side-by-side dollar table has filled in Google's blanks from somewhere.

Why there is no leaderboard here

The omission is deliberate. Benchmark positions between the major labs change on roughly a monthly cadence, and both vendors ship model revisions faster than any comparison page gets updated. A stale ranking is worse than none: it looks authoritative while being wrong in a specific direction.

The general shape has been stable for a while — both are frontier-class, the gap on ordinary application code is smaller than benchmark margins imply, and your prompt, context and harness account for more variance than the badge on the model. If you need a current ranking, read a leaderboard that updates continuously and check its date. Do not take one from a comparison article, including this one.

What is worth writing down is the material with dates attached: a retirement, a replacement, a product structure, a published tier. Those hold longer than a score does.

How to decide

Reasonable tie-breakers, in rough order of how often they turn out to be the real reason:

• You already pay one of them for something else. With a Claude subscription, Claude Code is not a new line item. On Google AI Pro or Ultra, Antigravity is where that entitlement now goes. This settles it more often than any technical argument.
• You want to try before paying. Antigravity's free tier is published, includes agent model access, and needs no subscription.
• You need a forecastable bill. Anthropic's figures are on a page. Google's, above the free tier, are not.
• You want one client across several labs. Antigravity ships Gemini, Claude and open-weight models in the same tool. Claude Code is Anthropic's models by design.
• You need source you can audit or fork. Neither is your answer, and that is the case for looking outside both.
• You are an enterprise Code Assist customer. Your access is explicitly unchanged, so the transition is something you can schedule rather than something done to you.

One piece of meta-advice for the whole category in 2026: check the date on whatever you read. Gemini CLI is not the only tool renamed this year — the same happened to Windsurf. Comparisons here have a half-life measured in months.

Sources

Every URL below was checked and read on 2026-07-31. Pricing and product structure change frequently here; verify before committing budget.

• Google Developers Blog — Transitioning Gemini CLI to Antigravity CLI (2026-05-19) — retirement dates and the enterprise carve-out, quoted above
• antigravity.google/pricing — the four tiers and the $0/month model list
• Antigravity docs — Plans · Getting started — quota language and version numbers
• claude.com/pricing (anthropic.com/pricing now redirects here) · Claude Code documentation
• google-gemini/gemini-cli · google-antigravity/antigravity-cli — contents and licence compared
• Google — Gemini Code Assist overview

Not affiliated with Google or Anthropic. If something here is wrong or has gone stale, tell us and we'll fix it.