If I'm not mistaken Codex is free until April 2nd with the previous generous rate limits (while paying customers get 2x).

Codex by a mile. Also, there's double rate limit until April. So you're paying 1 month for 2 months usage.

Antigravity (and Gemini in general) is not on par with the rest when it comes to agentic coding.

Between Codex and Claude, Codex will have much more generous limits for the same price, especially if you use top-of-the-line models (although for your task, Sonnet might actually be good enough).

Literally just find and replace

find and replace is step 1 that generates all the compile errors I want it to loop through

I'm wanting to do it on an entire programming language made in rust: https://github.com/uiua-lang/uiua

Because there are no float32 array languages in existence today

Why do you want a float32 array language? Anyway the free glm4.6 model that is opencode defaults to should be fine. Why pay for something to do this.

I want to use an array language for Real-time 3D. Float32 is faster for real-time calculations and can map memory directly to the GPU since 3D graphics runtimes are limited to float32.

All of them can do it but Codex has the least frustrating usage limits.

When using it in VSCode? The browser system running its own container seems like it would be the most demanding on their resources. The stand-alone client is Mac-only but I don't know if it makes a difference.

My goal is to do it within the usage I get from a $20 monthly plan.

You don't have to use their container thingy though, you can run Codex (CLI or VSCode, it doesn't matter) just fine in YOLO mode in your own local containers, or VMs, or however you want to isolate it.

Why would you use it in VSCode?

OpenAI are offering double the normal usage limits for Codex for two months. Go with them and do it in the terminal or the Mac OS codex app if you have a Mac.

It's different to use it in the terminal vs. vscode? Don't have a mac.

Sorry I wasn't aware it's available in vscode. Scratch my suggestion, then.

It is confusing especially when token efficiency is on the line.

Doesn't matter which one. All of them can do things like this now, given a good enough feedback loop. Which your problem has.