We're noticing a small percentage of calls using GPT 5.2 fail with an internal error during inference from OpenAI's side. We've reached out to the team and are closely monitoring the situation. In the meantime, we recommend switching to another model as we're seeing the degradation only on 5.2 currently.