Use lowest reasoning effort appropriate for a model (#2673)

* Use lowest reasoning effort appropriate for a model

* Address comments from Copilot

Author: pamelafox

app/backend/approaches/approach.py

@@ -127,18 +127,19 @@ def from_completion_usage(cls, usage: CompletionUsage) -> "TokenUsageProps":
 @dataclass
 class GPTReasoningModelSupport:
     streaming: bool
+    minimal_effort: bool
 
 
 class Approach(ABC):
     # List of GPT reasoning models support
     GPT_REASONING_MODELS = {
-        "o1": GPTReasoningModelSupport(streaming=False),
-        "o3": GPTReasoningModelSupport(streaming=True),
-        "o3-mini": GPTReasoningModelSupport(streaming=True),
-        "o4-mini": GPTReasoningModelSupport(streaming=True),
-        "gpt-5": GPTReasoningModelSupport(streaming=True),
-        "gpt-5-nano": GPTReasoningModelSupport(streaming=True),
-        "gpt-5-mini": GPTReasoningModelSupport(streaming=True),
+        "o1": GPTReasoningModelSupport(streaming=False, minimal_effort=False),
+        "o3": GPTReasoningModelSupport(streaming=True, minimal_effort=False),
+        "o3-mini": GPTReasoningModelSupport(streaming=True, minimal_effort=False),
+        "o4-mini": GPTReasoningModelSupport(streaming=True, minimal_effort=False),
+        "gpt-5": GPTReasoningModelSupport(streaming=True, minimal_effort=True),
+        "gpt-5-nano": GPTReasoningModelSupport(streaming=True, minimal_effort=True),
+        "gpt-5-mini": GPTReasoningModelSupport(streaming=True, minimal_effort=True),
     }
     # Set a higher token limit for GPT reasoning models
     RESPONSE_DEFAULT_TOKEN_LIMIT = 1024
@@ -412,6 +413,16 @@ def get_response_token_limit(self, model: str, default_limit: int) -> int:
 
         return default_limit
 
+    def get_lowest_reasoning_effort(self, model: str) -> ChatCompletionReasoningEffort:
+        """
+        Return the lowest valid reasoning_effort for the given model.
+        """
+        if model not in self.GPT_REASONING_MODELS:
+            return None
+        if self.GPT_REASONING_MODELS[model].minimal_effort:
+            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",

gpt5chat.txt

@@ -1,4 +1,4 @@
-{"delta": {"role": "assistant"}, "context": {"data_points": {"text": ["Benefit_Options-2.pdf: There is a whistleblower policy."], "images": null}, "thoughts": [{"title": "Prompt to generate search query", "description": [{"role": "system", "content": "Below is a history of the conversation so far, and a new question asked by the user that needs to be answered by searching in a knowledge base.\nYou have access to Azure AI Search index with 100's of documents.\nGenerate a search query based on the conversation and the new question.\nDo not include cited source filenames and document names e.g. info.txt or doc.pdf in the search query terms.\nDo not include any text inside [] or <<>> in the search query terms.\nDo not include any special characters like '+'.\nIf the question is not in English, translate the question to English before generating the search query.\nIf you cannot generate a search query, return just the number 0."}, {"role": "user", "content": "How did crypto do last year?"}, {"role": "assistant", "content": "Summarize Cryptocurrency Market Dynamics from last year"}, {"role": "user", "content": "What are my health plans?"}, {"role": "assistant", "content": "Show available health plans"}, {"role": "user", "content": "Generate search query for: What is the capital of France?"}], "props": {"model": "o3-mini", "deployment": "o3-mini", "reasoning_effort": "minimal", "token_usage": {"prompt_tokens": 23, "completion_tokens": 896, "reasoning_tokens": 384, "total_tokens": 919}}}, {"title": "Search using generated search query", "description": "capital of France", "props": {"use_semantic_captions": false, "use_semantic_ranker": false, "use_query_rewriting": false, "top": 3, "filter": null, "use_vector_search": false, "use_text_search": true}}, {"title": "Search results", "description": [{"id": "file-Benefit_Options_pdf-42656E656669745F4F7074696F6E732E706466-page-2", "content": "There is a whistleblower policy.", "category": null, "sourcepage": "Benefit_Options-2.pdf", "sourcefile": "Benefit_Options.pdf", "oids": null, "groups": null, "captions": [{"additional_properties": {}, "text": "Caption: A whistleblower policy.", "highlights": []}], "score": 0.03279569745063782, "reranker_score": 3.4577205181121826, "search_agent_query": null}], "props": null}, {"title": "Prompt to generate answer", "description": [{"role": "system", "content": "Assistant helps the company employees with their healthcare plan questions, and questions about the employee handbook. Be brief in your answers.\nAnswer ONLY with the facts listed in the list of sources below. If there isn't enough information below, say you don't know. Do not generate answers that don't use the sources below. If asking a clarifying question to the user would help, ask the question.\nIf the question is not in English, answer in the language used in the question.\nEach source has a name followed by colon and the actual information, always include the source name for each fact you use in the response. Use square brackets to reference the source, for example [info1.txt]. Don't combine sources, list each source separately, for example [info1.txt][info2.pdf]."}, {"role": "user", "content": "What is the capital of France?\n\nSources:\n\nBenefit_Options-2.pdf: There is a whistleblower policy."}], "props": {"model": "o3-mini", "deployment": "o3-mini", "reasoning_effort": null}}], "followup_questions": null}, "session_state": null}
+{"delta": {"role": "assistant"}, "context": {"data_points": {"text": ["Benefit_Options-2.pdf: There is a whistleblower policy."], "images": null}, "thoughts": [{"title": "Prompt to generate search query", "description": [{"role": "system", "content": "Below is a history of the conversation so far, and a new question asked by the user that needs to be answered by searching in a knowledge base.\nYou have access to Azure AI Search index with 100's of documents.\nGenerate a search query based on the conversation and the new question.\nDo not include cited source filenames and document names e.g. info.txt or doc.pdf in the search query terms.\nDo not include any text inside [] or <<>> in the search query terms.\nDo not include any special characters like '+'.\nIf the question is not in English, translate the question to English before generating the search query.\nIf you cannot generate a search query, return just the number 0."}, {"role": "user", "content": "How did crypto do last year?"}, {"role": "assistant", "content": "Summarize Cryptocurrency Market Dynamics from last year"}, {"role": "user", "content": "What are my health plans?"}, {"role": "assistant", "content": "Show available health plans"}, {"role": "user", "content": "Generate search query for: What is the capital of France?"}], "props": {"model": "o3-mini", "deployment": "o3-mini", "reasoning_effort": "low", "token_usage": {"prompt_tokens": 23, "completion_tokens": 896, "reasoning_tokens": 384, "total_tokens": 919}}}, {"title": "Search using generated search query", "description": "capital of France", "props": {"use_semantic_captions": false, "use_semantic_ranker": false, "use_query_rewriting": false, "top": 3, "filter": null, "use_vector_search": false, "use_text_search": true}}, {"title": "Search results", "description": [{"id": "file-Benefit_Options_pdf-42656E656669745F4F7074696F6E732E706466-page-2", "content": "There is a whistleblower policy.", "category": null, "sourcepage": "Benefit_Options-2.pdf", "sourcefile": "Benefit_Options.pdf", "oids": null, "groups": null, "captions": [{"additional_properties": {}, "text": "Caption: A whistleblower policy.", "highlights": []}], "score": 0.03279569745063782, "reranker_score": 3.4577205181121826, "search_agent_query": null}], "props": null}, {"title": "Prompt to generate answer", "description": [{"role": "system", "content": "Assistant helps the company employees with their healthcare plan questions, and questions about the employee handbook. Be brief in your answers.\nAnswer ONLY with the facts listed in the list of sources below. If there isn't enough information below, say you don't know. Do not generate answers that don't use the sources below. If asking a clarifying question to the user would help, ask the question.\nIf the question is not in English, answer in the language used in the question.\nEach source has a name followed by colon and the actual information, always include the source name for each fact you use in the response. Use square brackets to reference the source, for example [info1.txt]. Don't combine sources, list each source separately, for example [info1.txt][info2.pdf]."}, {"role": "user", "content": "What is the capital of France?\n\nSources:\n\nBenefit_Options-2.pdf: There is a whistleblower policy."}], "props": {"model": "o3-mini", "deployment": "o3-mini", "reasoning_effort": null}}], "followup_questions": null}, "session_state": null}
 {"delta": {"content": null, "role": "assistant"}}
 {"delta": {"content": "The capital of France is Paris. [Benefit_Options-2.pdf].", "role": null}}
-{"delta": {"role": "assistant"}, "context": {"data_points": {"text": ["Benefit_Options-2.pdf: There is a whistleblower policy."], "images": null}, "thoughts": [{"title": "Prompt to generate search query", "description": [{"role": "system", "content": "Below is a history of the conversation so far, and a new question asked by the user that needs to be answered by searching in a knowledge base.\nYou have access to Azure AI Search index with 100's of documents.\nGenerate a search query based on the conversation and the new question.\nDo not include cited source filenames and document names e.g. info.txt or doc.pdf in the search query terms.\nDo not include any text inside [] or <<>> in the search query terms.\nDo not include any special characters like '+'.\nIf the question is not in English, translate the question to English before generating the search query.\nIf you cannot generate a search query, return just the number 0."}, {"role": "user", "content": "How did crypto do last year?"}, {"role": "assistant", "content": "Summarize Cryptocurrency Market Dynamics from last year"}, {"role": "user", "content": "What are my health plans?"}, {"role": "assistant", "content": "Show available health plans"}, {"role": "user", "content": "Generate search query for: What is the capital of France?"}], "props": {"model": "o3-mini", "deployment": "o3-mini", "reasoning_effort": "minimal", "token_usage": {"prompt_tokens": 23, "completion_tokens": 896, "reasoning_tokens": 384, "total_tokens": 919}}}, {"title": "Search using generated search query", "description": "capital of France", "props": {"use_semantic_captions": false, "use_semantic_ranker": false, "use_query_rewriting": false, "top": 3, "filter": null, "use_vector_search": false, "use_text_search": true}}, {"title": "Search results", "description": [{"id": "file-Benefit_Options_pdf-42656E656669745F4F7074696F6E732E706466-page-2", "content": "There is a whistleblower policy.", "category": null, "sourcepage": "Benefit_Options-2.pdf", "sourcefile": "Benefit_Options.pdf", "oids": null, "groups": null, "captions": [{"additional_properties": {}, "text": "Caption: A whistleblower policy.", "highlights": []}], "score": 0.03279569745063782, "reranker_score": 3.4577205181121826, "search_agent_query": null}], "props": null}, {"title": "Prompt to generate answer", "description": [{"role": "system", "content": "Assistant helps the company employees with their healthcare plan questions, and questions about the employee handbook. Be brief in your answers.\nAnswer ONLY with the facts listed in the list of sources below. If there isn't enough information below, say you don't know. Do not generate answers that don't use the sources below. If asking a clarifying question to the user would help, ask the question.\nIf the question is not in English, answer in the language used in the question.\nEach source has a name followed by colon and the actual information, always include the source name for each fact you use in the response. Use square brackets to reference the source, for example [info1.txt]. Don't combine sources, list each source separately, for example [info1.txt][info2.pdf]."}, {"role": "user", "content": "What is the capital of France?\n\nSources:\n\nBenefit_Options-2.pdf: There is a whistleblower policy."}], "props": {"model": "o3-mini", "deployment": "o3-mini", "reasoning_effort": null, "token_usage": {"prompt_tokens": 23, "completion_tokens": 896, "reasoning_tokens": 384, "total_tokens": 919}}}], "followup_questions": null}, "session_state": null}
+{"delta": {"role": "assistant"}, "context": {"data_points": {"text": ["Benefit_Options-2.pdf: There is a whistleblower policy."], "images": null}, "thoughts": [{"title": "Prompt to generate search query", "description": [{"role": "system", "content": "Below is a history of the conversation so far, and a new question asked by the user that needs to be answered by searching in a knowledge base.\nYou have access to Azure AI Search index with 100's of documents.\nGenerate a search query based on the conversation and the new question.\nDo not include cited source filenames and document names e.g. info.txt or doc.pdf in the search query terms.\nDo not include any text inside [] or <<>> in the search query terms.\nDo not include any special characters like '+'.\nIf the question is not in English, translate the question to English before generating the search query.\nIf you cannot generate a search query, return just the number 0."}, {"role": "user", "content": "How did crypto do last year?"}, {"role": "assistant", "content": "Summarize Cryptocurrency Market Dynamics from last year"}, {"role": "user", "content": "What are my health plans?"}, {"role": "assistant", "content": "Show available health plans"}, {"role": "user", "content": "Generate search query for: What is the capital of France?"}], "props": {"model": "o3-mini", "deployment": "o3-mini", "reasoning_effort": "low", "token_usage": {"prompt_tokens": 23, "completion_tokens": 896, "reasoning_tokens": 384, "total_tokens": 919}}}, {"title": "Search using generated search query", "description": "capital of France", "props": {"use_semantic_captions": false, "use_semantic_ranker": false, "use_query_rewriting": false, "top": 3, "filter": null, "use_vector_search": false, "use_text_search": true}}, {"title": "Search results", "description": [{"id": "file-Benefit_Options_pdf-42656E656669745F4F7074696F6E732E706466-page-2", "content": "There is a whistleblower policy.", "category": null, "sourcepage": "Benefit_Options-2.pdf", "sourcefile": "Benefit_Options.pdf", "oids": null, "groups": null, "captions": [{"additional_properties": {}, "text": "Caption: A whistleblower policy.", "highlights": []}], "score": 0.03279569745063782, "reranker_score": 3.4577205181121826, "search_agent_query": null}], "props": null}, {"title": "Prompt to generate answer", "description": [{"role": "system", "content": "Assistant helps the company employees with their healthcare plan questions, and questions about the employee handbook. Be brief in your answers.\nAnswer ONLY with the facts listed in the list of sources below. If there isn't enough information below, say you don't know. Do not generate answers that don't use the sources below. If asking a clarifying question to the user would help, ask the question.\nIf the question is not in English, answer in the language used in the question.\nEach source has a name followed by colon and the actual information, always include the source name for each fact you use in the response. Use square brackets to reference the source, for example [info1.txt]. Don't combine sources, list each source separately, for example [info1.txt][info2.pdf]."}, {"role": "user", "content": "What is the capital of France?\n\nSources:\n\nBenefit_Options-2.pdf: There is a whistleblower policy."}], "props": {"model": "o3-mini", "deployment": "o3-mini", "reasoning_effort": null, "token_usage": {"prompt_tokens": 23, "completion_tokens": 896, "reasoning_tokens": 384, "total_tokens": 919}}}], "followup_questions": null}, "session_state": null}