Antigravity vs Claude Code vs Codex in Late 2026

My Medium friends can read this over on Medium.

I ranked these three in June. Not long ago. But AI is moving fast. And it isn’t kind to slow adapters.

In June, my ranking was Claude Code first, Codex right behind it, Antigravity back on my Mac after I had uninstalled it once already.

Three months later, Claude Code is still the one I open first. But it’s not really the clear winner anymore.

Changes since June

Codex switched its default model on September 4. The CLI runs GPT-6 Astra now when you don't configure anything, listed under version 0.153.4 in the changelog.

And Codex has gotten enormously great. At many tasks. With a good price tag attached.

On the Claude side there is Opus 5, and Fable 5.1 comes with Pro and Max at half the weekly limits. Still freaking expensive. And the capabilities… aren’t what they used to be in comparison to the others.

Antigravity is a family of products instead of one app. The 2.0 desktop app, the IDE, a CLI, an SDK. Version 2.14 shipped on September 15.

So if you read a comparison of these tools that was written in the spring, it is describing models that are not the default anywhere anymore. Including mine.

Everything is a CLI now

Antigravity started as a fork of VS Code. A whole IDE, with Gemini in it, sold on the idea that the editor is where the agent belongs.

Then it shipped a command line tool, and the command line tool got the best new feature of the three: Remote Control. You register it as a background service and drive a session on your Mac from a browser somewhere else, with push notifications when it needs you. That landed on September 10, per the Antigravity changelog.

Claude Code has been a terminal tool from day one. Codex too. Google built the IDE and then came back to the terminal anyway.

Claude Code

It holds a big, messy project in its head better than the other two. For now. But the margin is tiny.

What changed is how often I have to go back over its work. Since Opus 5 I catch more mistakes than I used to. A lot more. Small ones and bigger ones. But they add up over a day, and they cost more than they should.

It is also the only one of the three you cannot try without paying.

Codex

Codex in the terminal is fast, and close enough to Claude Code that it stays in rotation. In June I used it mostly as the reviewer of the Claude Code work.

Now I use it more and more. And I will actually set up a proper paid Codex plan with all the skills, plugins, connectors and such at work to give it a real go side-by-side against the Claude workflow we have there.

I suspect we will use both from then on. For different tasks.

Antigravity

Antigravity has a free tier. Zero dollars, basic weekly rate limits, and Gemini, Claude and GPT models in the same window, per Google's own pricing page. Not the newest Claude or OpenAI models, but it’s free.

Gemini isn't great for coding, in my eyes. Still. For image and video it’s pretty good. But ChatGPT is too. It’s gotten much better.

The models are not what Antigravity sells, though. It sells everything around the code. The browser, the Google suite, the design side, all in one window, which is still the reason I keep looking at it even though I wrote it off once.

But at this moment, Google is really not on top of the AI race. And not close either.

Money

Claude Code: 20 dollars a month on Pro, 17 if you pay for a year. Max starts at 100. But for the most capable models, it’s not uncommon that you need the 20x max plan for $200. Claude is freaking expensive.

Codex: in the free ChatGPT plan, 8 dollars on Go, 20 on Plus, from 100 on Pro. Plus gets you somewhere between 5 and 45 Astra messages per five hours, depending on the size of the task. Codex has much better pricing than Claude while not being behind on capabilities anymore.

Antigravity: free, with weekly limits. More generous limits come with Google AI Pro, €21.99 a month where I live. But it’s just not on par with the other two.

Three months ago I would have said take Claude Code and stop reading. Now I would ask what you do all day. If the answer is many tasks from coding, to websites, to writing, to images, to design, ChatGPT is now the clear winner. If the main thing you do is coding and design work, Claude is still the winner with the model setup and Claude Design.

Downsides, of course

Money. Again. Claude Code is the expensive one, and heavy use walks you toward the 200 dollar plan faster than you'd like.

Codex has really not big downsides anymore. It’s the best allrounder tool of the 3. And that says a lot. For me, the only downside is that the writing still sounds way more like AI than Claude does.

Antigravity… loses by a lot here. And Google overall. They need to catch up quickly.

To be fair, all three want your project on their servers. There is no version of this where the work stays on your machine. Not with these three, anyway.

The Bottom Line

In June the answer to what the best AI tool was: Claude Code for me. Now it depends on what you do all day, and on what you're willing to pay for a tool that is wrong sometimes (too many times for the price).

But if you have never paid for a coding agent and you're not sure you want one, Antigravity costs nothing and is good enough to find out. That’s really the only thing going for Antigravity right now.

If you want the best allrounder, pick ChatGPT. On a cheaper paid plan. It’s great.

If you're not already, follow me on Medium.