[Bug] Model downgrade mid-task causing loss of progress and context

Bug Description
you really need to fix fable, I can almost understand you blocking code related to genetic algorithms, which although frustrating, is tangential to wet lab work, even though I specificall stated this is only as related to software genetic algorithms. I understand less so on my research on Dendsovian vs Neaderthan physical properties. But I can't grasp at all why you could flag me trying to fix a web page that Opus built, that fails to log in, of course this is security related, but this is the simplest type of an OAuth debugging excercise, rediculous to flag that when any 32B open source model could answer the question. What is the point of having a model that won't do nearly anything that is asked of it, and then drops back to a lesser model, half way through the work? This is pretty much wasting all of the work that came before it at a high token rate, when it is switched to Opus, why even start if you can't deliver? And if you do "bail" on answering a question, you should refund all tokens spent on the work. When are you going to fix all of this?

Environment Info

• Platform: linux
• Terminal: gnome-terminal
• Version: 2.1.223
• Feedback ID: 7b3c6033-cea4-4c9f-b00b-373b1d4a4f3e

Errors

[]