Use lowest reasoning effort appropriate for a model

Author: pamelafox

app/backend/approaches/approach.py

@@ -412,6 +412,17 @@ def get_response_token_limit(self, model: str, default_limit: int) -> int:
 
         return default_limit
 
+    def get_lowest_reasoning_effort(self, model: str) -> Optional[ChatCompletionReasoningEffort]:
+        """
+        Return the lowest valid reasoning_effort for the given model.
+
+        Currently, "minimal" is only valid for the gpt-5 family.
+        For other reasoning models, return "low".
+        """
+        if model in {"gpt-5", "gpt-5-mini"}:
+            return "minimal"
+        return "low"
+
     def create_chat_completion(
         self,
         chatgpt_deployment: Optional[str],

app/backend/approaches/chatreadretrieveread.py

@@ -162,7 +162,7 @@ async def run_search_approach(
                 ),  # Setting too low risks malformed JSON, setting too high may affect performance
                 temperature=0.0,  # Minimize creativity for search query generation
                 tools=tools,
-                reasoning_effort="minimal",  # Minimize reasoning for search query generation
+                reasoning_effort=self.get_lowest_reasoning_effort(self.chatgpt_model),
             ),
         )
 
@@ -202,7 +202,7 @@ async def run_search_approach(
                     model=self.chatgpt_model,
                     deployment=self.chatgpt_deployment,
                     usage=chat_completion.usage,
-                    reasoning_effort="minimal",
+                    reasoning_effort=self.get_lowest_reasoning_effort(self.chatgpt_model),
                 ),
                 ThoughtStep(
                     "Search using generated search query",