My new rank for how I'm using models: - GPT 5.4 Mini for edits and dumb work, it's fast and works most times. If it fails, I can fix it, and in the second round, I get the cache discount. I use it the same way Cursor users use Composer. - GPT 5.3 Codex for when I start the work in Codex, most times I start the work in ChatGPT and copy the output to my worker model. - GPT 5.4 when I need a bit more power. 5.4 is a bit more expensive than 5.3 Codex, so I gravitate to Codex and Mini more now. Gemini CLI team is signalling that the free lunch is over, so I'll need to get a Pro account, meh, it's whatever. I did like 3 Flash as my worker, but on free accounts, it's not worth it anymore, you get maybe 4 big tasks in before you're out. I think the token kindness has moved over to paid users; I get why. Gemini had the most token abuse. We have zero Claude in my workflow 😭 Evil has lost. I repeat, Evil has lost. This might change when new Ant models arrive.

I am a Plus user for GPT. I am a Workspace Starter user, which counts for nothing, same as a free user.

@Presidentlin As a pro user of gemini the rate limits are really bad on antigravity

@Presidentlin I think nobody doing serious work is using Claude for anything these days. When it's not unavailable it's just bad compared to Codex, which was not the case not too long ago.

@Presidentlin I really recommend using Opus! As a reviewer it is just great. @doodlestein goes one step further and says you can even use Opus to write code if you have another Opus review/feedback it. But obv that’s the polar opposite to a free plan 😅

Pick which twin will win! You could score a share of $2 million in Bonus Bets.