BadRequestError: temperature=0.0 breaks evaluation with reasoning models (gpt-5.2, etc.)

[shubchat]

temperature=0.0 default breaks evaluation with reasoning models (gpt-5.2, etc.)

Summary

Running tau2 run with any reasoning model fails immediately with a 400 BadRequestError
because temperature=0.0 is hardcoded into the default LLM args.

Steps to Reproduce

tau2 run --domain airline \
  --agent-llm o4-mini \
  --user-llm o4-mini \
  --num-trials 1 --num-tasks 1

Error

  litellm.BadRequestError: AzureException BadRequestError -
  Unsupported value: 'temperature' does not support 0.0 with this model.
  Only the default (1) value is supported.