Output is much worse in a custom GPT vs using the API

[median rover]

Using the exact same instructions, I get consistently good results when using the API. But when the same instructions are used to create a custom GPT, it misses obvious things and is generally really bad compared to the API. I don't know if this is a bug or if there is a reason for this. OR - if there is a better way to do this? (basically, want to be able to run these custom instructions in my browser instead of getting into my coding environment and running a script via API each time)