Opus 5.5 medium spun up 12 Opus 5.5 medium subagents to read 87 transcripts and burned my entire 5-hour limit in 5 minutes lmao
retained text
https://preview.redd.it/u3kbdnm15nwh1.png?width=1102&format=png&auto=webp&s=3449c4f9ffe3192ba26962145246439b559e53f8 Wanted to use Opus 5.5 (medium) to restructure a big folder. It quickly found a subfolder with 87 .txt files, almost 5 mb's of conference transcripts and decided it was *essential* to create 12 subagents — also Opus 5.5 medium — to mine them. Burned through my entire 5-hour limit in 5 minutes. I’m not even mad. I had a laughing fit when I saw what happened. But I found out there’s apparently no way to set standard/default models for subagents? That feels like a feature that needs to be added soon. It seems natural. I can’t think of many agent tasks that actually need Opus-level capabilities. Has anyone found a workaround, or is this just a known limitation in the Claude desktop app? edit: thanks to everyone who replied! Aparently there *is* a way, via the [`claude.md`](http://claude.md) file. I just had no idea that existed. Please excuse me, I don’t come from a programming/compsci background at all. But let’s be honest: this should be a dedicated setting, right? The way AI is evolving and the use of agents is becoming more common, especially after the release of Opus 5.5, it feels like a pretty natural thing to expose in the UI. I can’t think of many agent tasks that actually need Opus-level capabilities, and accidentally spawning a dozen Opus subagents is a very expensive way to learn that.
document fetched 2026-10-21 00:21 · canonical https://reddit.com/r/ClaudeAI/comments/1xavke4/opus_55_medium_spun_up_12_opus_55_medium