Test claims Argon burned 4M tokens without finishing a coding task
A user reports Gemini 4 Argon used ~4M tokens in Antigravity without finishing a coding task, while Claude Opus 5.5 used ~240K tokens and passed all 90 tests.
I JUST FOUND OUT WHY PEOPLE HATE GEMINI 4 ARGON.... I thought people were exaggerating… So I decided to test it myself. I gave Argon a simple coding task that took around 11 minutes It burned through ~4 MILLION tokens 😭 So I gave the exact same task to Claude Opus 5.5 Medium ~240K read tokens It completed the task and passed all 90 final verification tests Then I thought maybe Antigravity was the problem… So I added $10 of Gemini CLI credits and tried again Argon burned through the entire $10 Still couldn’t finish the task 💀 And people still tell me Argon is “cheaper” than other frontier models 👀 So what am I missing? Different harness? Better context management? A trick to stop repeated file reads? If you use Argon daily, please tell me what I’m doing wrong 😭