An Indie Developer's Honest Verdict: Claude's Top Model Burned Over 25% of Session Tokens on a Bug Fix -- Then Opus Nailed It in One Shot

This indie game developer says he's tired of being asked, almost every single stream on Twitch, whether he's coding by hand and why he doesn't just use Claude Code. He explains that he codes by hand because he's mid-way through a project with a complicated structure and wants to keep a full grasp of how everything works. Viewers sometimes accuse him of being 'a robot racist' for it -- so he decided it was time to actually put it to the test himself.

Handing two real bugs from his game to the top-tier Fable model alone burned through more than 25% of the session's tokens, he recalls, which raised real doubts about its practicality. Out of curiosity, he reproduced the same bug and switched the model to Opus -- and despite Fable supposedly being the smarter model, Opus delivered a perfect fix in one shot, restoring a check that had been deleted. He also tried Sonnet, which failed entirely, scattering unrelated changes through the code without ever actually resolving the issue.

His overall conclusion: better than expected, but not fully reliable. He won't use it to generate the main code for his own project, but for hunting down bugs -- something that can eat up half an hour of searching for the exact cause -- it can save a lot of time and frustration. Starting a brand-new project entirely by hand from now on would feel like 'going back to the stone age,' he admits, while also warning that leaning on it too much risks losing your grip on your own code to the point you can't add anything without asking Claude first. Used carefully, he concludes, it's a genuine productivity multiplier.