fix gpt-5

Author: xings19

config/agent_gaia-validation-gpt5.yaml

@@ -8,16 +8,16 @@ main_agent:
   prompt_class: MainAgentPrompt_GAIA
   llm: 
     provider_class: "GPT5OpenAIClient"
-    model_name: "gpt-5-2025-08-07"
+    model_name: "gpt-5"
     async_client: true
     temperature: 1.0
     top_p: 1.0
     min_p: 0.0
     top_k: -1
     max_tokens: 128000
     reasoning_effort: "high"
-    openrouter_api_key: "${oc.env:OPENAI_API_KEY,???}"
-    openrouter_base_url: "${oc.env:OPENAI_BASE_URL,https://api.openai.com/v1}"
+    openai_api_key: "${oc.env:OPENAI_API_KEY,???}"
+    openai_base_url: "${oc.env:OPENAI_BASE_URL,https://api.openai.com/v1}"
     openrouter_provider: ""
     disable_cache_control: true
     keep_tool_result: -1
@@ -30,11 +30,13 @@ main_agent:
   max_tool_calls_per_turn: 10  # Maximum number of tool calls per turn
 
   input_process:
-    o3_hint: true
+    hint_generation: true
+    hint_llm_base_url: "${oc.env:HINT_LLM_BASE_URL,https://api.openai.com/v1}"
   output_process:
-    o3_final_answer: true
+    final_answer_extraction: true
+    final_answer_llm_base_url: "${oc.env:FINAL_ANSWER_LLM_BASE_URL,https://api.openai.com/v1}"
 
-  openai_api_key: "${oc.env:OPENAI_API_KEY,???}" # used for o3 hints and final answer extraction
+  openai_api_key: "${oc.env:OPENAI_API_KEY,???}" # used for hint generation and final answer extraction
   add_message_id: true
   keep_tool_result: -1
   chinese_context: "${oc.env:CHINESE_CONTEXT,false}"
@@ -44,16 +46,17 @@ sub_agents:
   agent-worker:
     prompt_class: SubAgentWorkerPrompt
     llm: 
-      provider_class: "ClaudeOpenRouterClient"
-      model_name: "gpt-5-2025-08-07"
+      provider_class: "GPT5OpenAIClient"
+      model_name: "gpt-5"
       async_client: true
       temperature: 1.0
       top_p: 1.0
       min_p: 0.0
       top_k: -1
       max_tokens: 128000
-      openrouter_api_key: "${oc.env:OPENAI_API_KEY,???}"
-      openrouter_base_url: "${oc.env:OPENAI_BASE_URL,https://api.openai.com/v1}"
+      reasoning_effort: "medium"
+      openai_api_key: "${oc.env:OPENAI_API_KEY,???}"
+      openai_base_url: "${oc.env:OPENAI_BASE_URL,https://api.openai.com/v1}"
       openrouter_provider: ""
       disable_cache_control: true
       keep_tool_result: -1

src/llm/providers/gpt5_openai_client.py

@@ -37,14 +37,14 @@ def _create_client(self, config: DictConfig):
         """Create configured OpenAI client"""
         if self.async_client:
             return AsyncOpenAI(
-                api_key=self.cfg.llm.openrouter_api_key,
-                base_url=self.cfg.llm.openrouter_base_url,
+                api_key=self.cfg.llm.openai_api_key,
+                base_url=self.cfg.llm.openai_base_url,
                 timeout=1800,
             )
         else:
             return OpenAI(
-                api_key=self.cfg.llm.openrouter_api_key,
-                base_url=self.cfg.llm.openrouter_base_url,
+                api_key=self.cfg.llm.openai_api_key,
+                base_url=self.cfg.llm.openai_base_url,
                 timeout=1800,
             )
 

src/llm/providers/gpt_openai_client.py

@@ -18,7 +18,7 @@
 
 LOGGER_LEVEL = os.getenv("LOGGER_LEVEL", "INFO")
 # OPENAI reasoning models only support temperature=1
-OPENAI_REASONING_MODEL_SET = set(["o1", "o3", "o3-mini", "o4-mini"])
+OPENAI_REASONING_MODEL_SET = set(["o1", "o3", "o3-mini", "o4-mini", "gpt-5", "gpt-5-2025-08-07"])
 
 logger = bootstrap_logger(level=LOGGER_LEVEL)
 
@@ -29,13 +29,15 @@ def _create_client(self, config: DictConfig):
         """Create configured OpenAI client"""
         if self.async_client:
             return AsyncOpenAI(
-                api_key=config.env.openai_api_key,
-                base_url=config.env.openai_base_url,
+                api_key=self.cfg.llm.openai_api_key,
+                base_url=self.cfg.llm.openai_base_url,
+                timeout=1800,
             )
         else:
             return OpenAI(
-                api_key=config.env.openai_api_key,
-                base_url=config.env.openai_base_url,
+                api_key=self.cfg.llm.openai_api_key,
+                base_url=self.cfg.llm.openai_base_url,
+                timeout=1800,
             )
 
     @retry(wait=wait_fixed(10), stop=stop_after_attempt(5))
@@ -58,6 +60,7 @@ async def _create_message(
             or self.model_name.startswith("o4")
             or self.model_name.startswith("gpt-4.1")
             or self.model_name.startswith("gpt-4o")
+            or self.model_name.startswith("gpt-5")
         )
         logger.debug(f" Calling LLM ({'async' if self.async_client else 'sync'})")
         # put the system prompt in the first message since OpenAI API does not support system prompt in
@@ -88,21 +91,28 @@ async def _create_message(
         tool_list = await self.convert_tool_definition_to_tool_call(tools_definitions)
 
         try:
-            # Set temperature=1 for reasoning models
-            temperature = (
-                1.0
-                if self.model_name in OPENAI_REASONING_MODEL_SET
-                else self.temperature
-            )
-
-            params = {
-                "model": self.model_name,
-                "temperature": temperature,
-                "max_completion_tokens": self.max_tokens,
-                "messages": messages_copy,
-                "tools": tool_list,
-                "stream": False,
-            }
+            # Set temperature and reasoning_effort for reasoning models
+            if self.model_name in OPENAI_REASONING_MODEL_SET:
+                temperature = 1.0
+                params = {
+                    "model": self.model_name,
+                    "temperature": temperature,
+                    "max_completion_tokens": self.max_tokens,
+                    "messages": messages_copy,
+                    "reasoning_effort": self.reasoning_effort,
+                    "tools": tool_list,
+                    "stream": False,
+                }
+            else:
+                temperature = self.temperature
+                params = {
+                    "model": self.model_name,
+                    "temperature": temperature,
+                    "max_completion_tokens": self.max_tokens,
+                    "messages": messages_copy,
+                    "tools": tool_list,
+                    "stream": False,
+                }
 
             if self.top_p != 1.0:
                 params["top_p"] = self.top_p