Hey, thanks for the report. This isn’t a bug, it’s the current behavior for request-based plans, including Free, without Max Mode. Subagents are forced to stay on the Composer family to control costs. So when you select GPT-5.5, it fails the check and falls back to Composer 2.5 Fast.
I agree the UX is off here. The subagent editor lets you pick GPT-5.5 even though it won’t apply on your plan, and there’s no warning. I’ll pass this along.
Allowing subagents to use a different model on request-based plans is a feature request that’s being tracked. No ETA yet.