Confirm this is an issue with the Python library and not an underlying OpenAI API

• This is an issue with the Python library

Describe the bug

In version 0.28 of the openai package, I used the following code:
message_construct = [{"role": "user", "content": prompt_message}] completion = openai.ChatCompletion.create( model="gpt-5-mini-2025-08-07", messages=message_construct, seed=2337541 )

When I upgraded to the latest openai package and used the new client interface:
client = OpenAI(api_key=OPENAI_KEY) message_construct = [{"role": "user", "content": prompt_message}] completion = client.chat.completions.create( model="gpt-5-mini-2025-08-07", messages=message_construct, seed=2337541 )

I found that the responses are very different, even though the prompt_message and seed are exactly the same in both cases.
Could you clarify why there is such a significant difference in the results between 0.28 and the latest version of the SDK? Is there any change in how requests are processed or in how the seed parameter is handled?

To Reproduce

message_construct = [{"role": "user", "content": prompt_message}] completion = openai.ChatCompletion.create( model="gpt-5-mini-2025-08-07", messages=message_construct, seed=2337541 )

=============

client = OpenAI(api_key=OPENAI_KEY) message_construct = [{"role": "user", "content": prompt_message}] completion = client.chat.completions.create( model="gpt-5-mini-2025-08-07", messages=message_construct, seed=2337541 )

Code snippets

OS

macOS

Python version

3.11

Library version

0.28 and latest