You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
{{ message }}
Repository navigation
Astra reasoning level frequently reset to Medium #44668
Tbh, medium seems to offer nearly the same results as higher efforts and costs around half the usage as extra high, check codexradar.com for details. It will be slower to complete the task compared to extra high, but from my personal experience the usage cost ends up being noticeably less for nearly the same end result. If you can deal with a little longer dev time, id consider using medium for most high level tasks. But yeah, I agree with you it shouldn't just change on its own.
This is a deal breaker. Constant unexpected switching to medium starts writing low-quality code that ruins all the efforts. There isn't even a time to commit the code so that it can be rolled back.
It wasn't a problem with Astra before, this seems to be specifically turned on to reduce the number of people using xhigh reasoning, so that the load on servers is reduced.
Making the product unusable for users is not a viable way to fix the server load issue.
Activity
github-actions commentedon Sep 10, 2026
Potential duplicates detected. Please review them and close your issue if it is a duplicate.
Powered by Codex Action
digitdash commentedon Sep 10, 2026
Tbh, medium seems to offer nearly the same results as higher efforts and costs around half the usage as extra high, check codexradar.com for details. It will be slower to complete the task compared to extra high, but from my personal experience the usage cost ends up being noticeably less for nearly the same end result. If you can deal with a little longer dev time, id consider using medium for most high level tasks. But yeah, I agree with you it shouldn't just change on its own.
reverofevil commentedon Sep 11, 2026
This is a deal breaker. Constant unexpected switching to medium starts writing low-quality code that ruins all the efforts. There isn't even a time to commit the code so that it can be rolled back.
It wasn't a problem with Astra before, this seems to be specifically turned on to reduce the number of people using xhigh reasoning, so that the load on servers is reduced.
Making the product unusable for users is not a viable way to fix the server load issue.