Hello, over the past few weeks I have noticed assistant outputs in my voice agents with terms such as: "json" or "final analysis to assistant". For context my voice assistants are healthcare characters so their output is only meant to be conversational human text. Been using VAPI for 18months now in production and first time we are seeing this. Prompts remain unchanged. Has anyone else experienced this? Unsure if a Vapi issue or an OpenAI LLM issue?

Hi there, Thank you for your message. Our team is currently out of the office. We operate Monday through Friday, from 9:00 AM to 8:00 PM Pacific Standard Time (PST). We’ll get back to you as soon as possible during our normal business hours. If your message is urgent, please mark it accordingly or include “URGENT” in the subject line, and we’ll do our best to respond promptly. Warm regards, Vapi Customer Support Team

That sounds more like an output formatting leak than a prompt issue, especially if terms like “json” or “final analysis” are showing in live calls. I’ve seen this happen from structured outputs, tool calling, or recent model changes leaking into assistant responses. Worth checking your parser, response settings, and any recent Vapi/OpenAI model updates. Happy to help you trace it directly are you using function/tool calling in the agent? @TopsyMustang

Hi @Matt , nil function/tool calling. Only change was a swap from gpt-4.1 to gpt-5.4...

That change is the culprit. They way to prompt different models changes ever so slightly between minor upgrades like 4o to 4.1, but changes significantly when making a jump over major versions like from 4 to 5. 4.1 is excellent for most use cases, but if 5.4 is a necessity I'd recommend adjusting the prompt to suit the model better.

Ok thanks @Ace K