Merge branch 'main' into openrouter_support

browser_use/agent/views.py

@@ -193,24 +193,25 @@ class AgentOutput(BaseModel):
 	def type_with_custom_actions_no_thinking(custom_actions: type[ActionModel]) -> type[AgentOutput]:
 		"""Extend actions with custom actions and exclude thinking field"""
 
-		# Create a base model without thinking, but inheriting from AgentOutput
-		# Override only the fields we need to change
-		model_ = create_model(
+		class AgentOutputNoThinking(AgentOutput):
+			@classmethod
+			def model_json_schema(cls, **kwargs):
+				schema = super().model_json_schema(**kwargs)
+				del schema['properties']['thinking']
+				return schema
+
+		model = create_model(
 			'AgentOutput',
-			__base__=AgentOutput,
-			thinking=(
-				type(None),  # type: ignore
-				Field(default=None, exclude=True),
-			),  # Exclude thinking from schema
+			__base__=AgentOutputNoThinking,
 			action=(
 				list[custom_actions],  # type: ignore
 				Field(..., description='List of actions to execute', json_schema_extra={'min_items': 1}),
 			),
-			__module__=AgentOutput.__module__,
+			__module__=AgentOutputNoThinking.__module__,
 		)
 
-		model_.__doc__ = 'AgentOutput model with custom actions'
-		return model_
+		model.__doc__ = 'AgentOutput model with custom actions'
+		return model
 
 
 class AgentHistory(BaseModel):

browser_use/llm/openai/chat.py

@@ -96,8 +96,15 @@ class ChatOpenAI(BaseChatModel):
 		return str(self.model)
 
 	def _get_usage(self, response: ChatCompletion) -> ChatInvokeUsage | None:
-		usage = (
-			ChatInvokeUsage(
+		if response.usage is not None:
+			completion_tokens = response.usage.completion_tokens
+			completion_token_details = response.usage.completion_tokens_details
+			if completion_token_details is not None:
+				reasoning_tokens = completion_token_details.reasoning_tokens
+				if reasoning_tokens is not None:
+					completion_tokens += reasoning_tokens
+
+			usage = ChatInvokeUsage(
 				prompt_tokens=response.usage.prompt_tokens,
 				prompt_cached_tokens=response.usage.prompt_tokens_details.cached_tokens
 				if response.usage.prompt_tokens_details is not None
@@ -105,12 +112,12 @@ class ChatOpenAI(BaseChatModel):
 				prompt_cache_creation_tokens=None,
 				prompt_image_tokens=None,
 				# Completion
-				completion_tokens=response.usage.completion_tokens,
+				completion_tokens=completion_tokens,
 				total_tokens=response.usage.total_tokens,
 			)
-			if response.usage is not None
-			else None
-		)
+		else:
+			usage = None
+
 		return usage
 
 	@overload