still fails to give any indication, warning, or otherwise show that the model is changed/locked to Opus 4.8
I noticed random usage (~20m tokens) of Opus 4.8 despite never selecting it myself (almost always on grok 4.6 high). After digging, it matches the above ticket from 1.5 months ago.
I’d have thought by this point Grok 4.6 would be more capable than Opus 4.8, while being cheaper per token, and significantly cheaper when using Cursor subscription usage.
As a side note, it was pretty confusing to see my usage in UTC time (took a while for me to realise), while my runs/automations show local time.
Steps to Reproduce
Setup cloud agent environment, in IDE or webapp.
Expected Behavior
best case scenario, allow grok 4.6 high for env setup
worst case scenario, show/tell user that Opus 4.8 is locked to the env setup and the conversation thereafter.
Hi @jnotts Thanks for the post! You’re right that environment setup still runs on a fixed model (currently Opus 4.8), regardless of the model you have selected. We pin it for setup reliability, and your first few setup runs are free. After that, any further messages in that same setup chat are charged normally, and they keep running on the setup model.
Looking at your account, that is what happened here. The Opus 4.8 usage you saw came from one setup session on Sep 2 where the setup itself and the first few follow-ups were free, but the conversation continued for another 14 follow-ups in that setup chat, and those were billed as Opus 4.8.
To avoid this going forward: once the environment finishes setting up and is saved, start a new cloud agent on that environment (rather than continuing in the setup chat) and pick Grok 4.6 High there. That agent runs on the model you choose.