Legacy completions and new chat completions returning exact same text every time

We have an application that generates property description for users. Every time user asks for another generation, we invoke the API with gpt-3.5-turbo-instruct. It was working great until 4 days back. Now it returns exact same text back every time. We don’t pass any seed to get back same result. We tested with both legacy completions and new chat completions using same 3.5 turbo instruct model. We want different results every time. Has anything changed? Thanks.