Hi @Artemonim
This is a known display bug. The correct model and reasoning effort are being used for your subagents (your usage log confirms gpt-5.4-high and gpt-5.4-xhigh were actually used). The preview card just shows the wrong effort level.
I see you’ve reported similar display mismatches in your earlier threads too. Our team is aware and actively working on a fix for this. Thanks for your patience, and for continuing to flag these.