Are the playground and the API the same model?

I am troubled by an inexplicable phenomenon.
It may only be in the Japanese environment, but I am using a configuration with Assistant API + Function call where I extract keywords in the first step and pass them to a function.
It works correctly in the playground, and I have already checked for differences in input encoding and format.
However, when I use the API, it turns into Japanese that doesn’t exist.
For example, when I ask ‘Please tell me about prenatal health checkups’, it should extract ‘妊娠健康診査’ (prenatal health checkups), but instead, it gives me ‘姊媚偶察診染’, which is not real Japanese. In the playground, it correctly extracts ‘妊娠健康診査’.

I am at a dead end.
is there any solution or clue to solve this?"

I switched to the older model gpt-3.5-turbo-0613 (I’m not sure if it’s compatible with Assistant), and it started functioning as intended. I’ll report this to the OpenAI team for your reference