diff --git a/common/config/config.go b/common/config/config.go
index a9e56b07..e98a1129 100644
--- a/common/config/config.go
+++ b/common/config/config.go
@@ -176,4 +176,5 @@ var RelayProxy = env.String("RELAY_PROXY", "")
 var UserContentRequestProxy = env.String("USER_CONTENT_REQUEST_PROXY", "")
 var UserContentRequestTimeout = env.Int("USER_CONTENT_REQUEST_TIMEOUT", 30)
 
+// EnforceIncludeUsage is used to determine whether to include usage in the response
 var EnforceIncludeUsage = env.Bool("ENFORCE_INCLUDE_USAGE", false)
diff --git a/relay/adaptor/openai/adaptor.go b/relay/adaptor/openai/adaptor.go
index ea001bee..68bdaeaf 100644
--- a/relay/adaptor/openai/adaptor.go
+++ b/relay/adaptor/openai/adaptor.go
@@ -91,7 +91,6 @@ func (a *Adaptor) ConvertRequest(c *gin.Context, relayMode int, request *model.G
 		request.Temperature = &temperature // Only the default (1) value is supported
 
 		request.MaxTokens = 0
-
 		request.Messages = func(raw []model.Message) (filtered []model.Message) {
 			for i := range raw {
 				if raw[i].Role != "system" {
@@ -103,10 +102,10 @@ func (a *Adaptor) ConvertRequest(c *gin.Context, relayMode int, request *model.G
 		}(request.Messages)
 	}
 
-	if request.Stream && strings.HasPrefix(request.Model, "gpt-4o-audio") {
+	if request.Stream && strings.HasPrefix(request.Model, "gpt-4o-audio") && !config.EnforceIncludeUsage {
 		// TODO: Since it is not clear how to implement billing in stream mode,
 		// it is temporarily not supported
-		return nil, errors.New("stream mode is not supported for gpt-4o-audio")
+		return nil, errors.New("set ENFORCE_INCLUDE_USAGE=true to enable stream mode for gpt-4o-audio")
 	}
 
 	return request, nil