Claude and ChatGPT’s newest fashions are nitpicky and burning tokens. Here is the repair

Even worse, once I ask one of many fashions to evaluation the difficulty of one other, the reviewing mannequin invariably finds all types of issues with the primary mannequin’s work, resulting in but extra multi-step fixes.

All these fixes are taking a toll on my utilization limits. I’m a Claude Professional and ChatGPT Plus subscriber ($20/month every), and recently I’m blowing half my weekly allowance chasing bugs discovered by Claude Opus 5 and GPT-5.6.

I’m all for being thorough, however the nitpicky nature of the brand new Claude and GPT fashions is akin to a physique store mechanic who needs to exchange the hood of your automotive after recognizing a tiny nick within the paint. And this isn’t simply a difficulty for coders, by the best way—it’s an issue for anybody who wants assist with enhancing work proposals, reviewing Excel spreadsheets, or performing any variety of on a regular basis AI duties whereas on a token finances.

So I requested Claude Opus 5 for an “on a finances” immediate that takes a process and asks the AI to respect your utilization limits, separating actual issues from beauty ones and cutting down the scope to one thing that’s thoughtful of your time and pockets.

Naturally, Opus 5 spat out a reasonably prolonged six-point plan for the immediate, and once I requested GPT-5.6 Sol to evaluation it, it gave me a dozen detailed criticisms. Telling GPT-5.6 to return and comply with the directions within the authentic immediate, it shortly backtracked (“You’re proper—I gold-plated the evaluation”) and narrowed the listing to 3.

Right here’s the ultimate immediate that Opus 5 and GPT-5.6 Sol agreed upon, pared right down to 4 bullet factors from the unique six (and sure, it borrowed the “physique store” metaphor I initially gave it):