Antigravity vs Claude Code vs Codex
I praised Antigravity, then dropped it. The new version looked great, though.
My Medium friends can read this over on Medium.
I praised Antigravity, then dropped it. The new version looked great, though.
I have changed my mind about Antigravity twice now.
First I wrote that Antigravity beat Claude Code for the way I work. Then it started crashing and writing worse code, so I dropped it and went back to the terminal.
Now there is a new version. And it changed my mind again.
Here is where I land in 2026. Claude Code is the most reliable of the three. Codex is the fast terminal alternative that gets very close. And the new Antigravity finally works, now a lot like the Codex Mac app, far better than the version I uninstalled.
For most people building real things, Claude Code is still the safe default. But Antigravity earned a second look. Let me go through all three.
The Antigravity backstory
Antigravity is Google's coding tool. A fork of VS Code with Gemini built in, visual awareness, Chrome access, and the whole Google suite sitting next to your code.
The first version was exciting. Then it fell apart. Crashes mid-task. "Agent error" with no explanation. RAM usage that spun up the fans. Code that looked right and wasn't. So I uninstalled it and said so.
I did not expect to come back...
The new Antigravity
The new version is like a different product. Closer to the Codex Mac app from OpenAI than to the heavy IDE that kept crashing on me. Although, there still is the IDE version as well.
Antigravity 2 is calmer and more stable. It does the job without failing every twenty minutes.
I am not ready to call it the best of the three. Not close. But it works now. That alone is a big jump from version one. And it still brings what made the first one tempting, the Google suite right next to the code, image and video generation, Gemini and Claude models under one roof.
Claude Code
Claude Code is my default again, and has been for a while. It runs in the terminal, the place AI made cool again. I keep a stack of terminal tools next to it.
It is boring in the best way. It almost never crashes. It holds context across a big, messy project, and it makes fewer dumb mistakes than anything else I use. Opus 4.8 made it sharper. Fable would too if I had access...
It is also the most expensive. That is the catch...
Codex
Codex is pretty great. I am not an OpenAI fan. But Codex in the terminal is fast and accurate, close enough to Claude Code that I keep it in rotation. The Codex Mac app is nice too. Looks great, works well. Much better than the Claude Mac app. But I'd rather use the terminal right now.
For web apps, mobile, and writing, Codex performs almost at Claude Code's level. Sometimes faster. I mainly use Codex as the reviewer of the Claude Code work.
Antigravity vs Claude Code vs Codex
Side by side, for the way I work:
Claude Code wins on reliability and on understanding big projects. Codex wins on speed and is the cheaper terminal pick that still gets the logic right. The new Antigravity wins on everything around the code, the browser, the design, the Google tools, all in one window.
But for you, the call depends on what you do. If you ship serious code and want one thing to just work, Claude Code is still #1. If cost matters more, Codex or Antigravity. If you want the Google world, Antigravity is good again now.
Options
Each one costs you something.
Claude Code costs money, more than the others, a lot more if you use it heavily. Antigravity asks you to forgive a rough history and bet that the stability holds this time. Codex means trusting OpenAI, which not everyone wants to do.
The Bottom Line
Tools change, month to month. I keep changing with them.
Right now Claude Code is still the one I trust most with the hard stuff. But for the first time since I uninstalled it, Antigravity is back on my Mac. And I am not mad about it.