Use max_completion_tokens for newer OpenAI models

Newer models (5.x+) reject max_tokens and require
max_completion_tokens instead. Convert in _create().

Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>

Author: vthorsteinsson

queries/util/openai_gpt.py

@@ -82,6 +82,9 @@ def _create(cls, *args: Any, **kwargs: Any) -> ChatCompletion:
         """
         Creates a new completion while handling formatting and parsing.
         """
+        # Newer OpenAI models use max_completion_tokens instead of max_tokens
+        if "max_tokens" in kwargs:
+            kwargs["max_completion_tokens"] = kwargs.pop("max_tokens")
         return client.chat.completions.create(*args, model=MODEL, **kwargs)
 
     @classmethod