I am building a chatbot using OpenAI agent SDK. I recently migrated from GPT 4.1 to 5.1 with reasoning to ‘none’. and it seems that 5.1 does not follow the instruction/prompt properly and producing abnormal output.
For example, I have to tool to find customer from my database. and I gave instruction that if there is not customer found by the name provided then ask user if they want to create a new customer or not. But instead of doing that 5.1 generates fake customer names with the name given by the user.
But on 4.1 this instruction was working perfectly.
Anyone knows how to fix this?
1 Like