diff --git a/sgl-router/src/reasoning_parser/parsers/base.rs b/sgl-router/src/reasoning_parser/parsers/base.rs
index 0fd2818b9..2aa8846a9 100644
--- a/sgl-router/src/reasoning_parser/parsers/base.rs
+++ b/sgl-router/src/reasoning_parser/parsers/base.rs
@@ -187,6 +187,10 @@ impl ReasoningParser for BaseReasoningParser {
     fn model_type(&self) -> &str {
         &self.model_type
     }
+
+    fn is_in_reasoning(&self) -> bool {
+        self.in_reasoning
+    }
 }
 
 #[cfg(test)]
diff --git a/sgl-router/src/reasoning_parser/parsers/deepseek_r1.rs b/sgl-router/src/reasoning_parser/parsers/deepseek_r1.rs
index 62a7aadec..1bb2f4c48 100644
--- a/sgl-router/src/reasoning_parser/parsers/deepseek_r1.rs
+++ b/sgl-router/src/reasoning_parser/parsers/deepseek_r1.rs
@@ -55,6 +55,10 @@ impl ReasoningParser for DeepSeekR1Parser {
     fn model_type(&self) -> &str {
         self.base.model_type()
     }
+
+    fn is_in_reasoning(&self) -> bool {
+        self.base.is_in_reasoning()
+    }
 }
 
 #[cfg(test)]
diff --git a/sgl-router/src/reasoning_parser/parsers/glm45.rs b/sgl-router/src/reasoning_parser/parsers/glm45.rs
index e4e56723c..5b2778993 100644
--- a/sgl-router/src/reasoning_parser/parsers/glm45.rs
+++ b/sgl-router/src/reasoning_parser/parsers/glm45.rs
@@ -54,6 +54,10 @@ impl ReasoningParser for Glm45Parser {
     fn model_type(&self) -> &str {
         self.base.model_type()
     }
+
+    fn is_in_reasoning(&self) -> bool {
+        self.base.is_in_reasoning()
+    }
 }
 
 #[cfg(test)]
diff --git a/sgl-router/src/reasoning_parser/parsers/kimi.rs b/sgl-router/src/reasoning_parser/parsers/kimi.rs
index 3bc77b4f1..0095f94f0 100644
--- a/sgl-router/src/reasoning_parser/parsers/kimi.rs
+++ b/sgl-router/src/reasoning_parser/parsers/kimi.rs
@@ -54,6 +54,10 @@ impl ReasoningParser for KimiParser {
     fn model_type(&self) -> &str {
         self.base.model_type()
     }
+
+    fn is_in_reasoning(&self) -> bool {
+        self.base.is_in_reasoning()
+    }
 }
 
 #[cfg(test)]
diff --git a/sgl-router/src/reasoning_parser/parsers/qwen3.rs b/sgl-router/src/reasoning_parser/parsers/qwen3.rs
index 8c5ce9e8c..038e7db8d 100644
--- a/sgl-router/src/reasoning_parser/parsers/qwen3.rs
+++ b/sgl-router/src/reasoning_parser/parsers/qwen3.rs
@@ -55,6 +55,10 @@ impl ReasoningParser for Qwen3Parser {
     fn model_type(&self) -> &str {
         self.base.model_type()
     }
+
+    fn is_in_reasoning(&self) -> bool {
+        self.base.is_in_reasoning()
+    }
 }
 
 /// QwenThinking parser - variant that assumes reasoning from start.
@@ -106,6 +110,10 @@ impl ReasoningParser for QwenThinkingParser {
     fn model_type(&self) -> &str {
         self.base.model_type()
     }
+
+    fn is_in_reasoning(&self) -> bool {
+        self.base.is_in_reasoning()
+    }
 }
 
 #[cfg(test)]
diff --git a/sgl-router/src/reasoning_parser/parsers/step3.rs b/sgl-router/src/reasoning_parser/parsers/step3.rs
index cec0bcd15..155e340cc 100644
--- a/sgl-router/src/reasoning_parser/parsers/step3.rs
+++ b/sgl-router/src/reasoning_parser/parsers/step3.rs
@@ -54,6 +54,10 @@ impl ReasoningParser for Step3Parser {
     fn model_type(&self) -> &str {
         self.base.model_type()
     }
+
+    fn is_in_reasoning(&self) -> bool {
+        self.base.is_in_reasoning()
+    }
 }
 
 #[cfg(test)]
diff --git a/sgl-router/src/reasoning_parser/traits.rs b/sgl-router/src/reasoning_parser/traits.rs
index d435e6c23..c21e342f0 100644
--- a/sgl-router/src/reasoning_parser/traits.rs
+++ b/sgl-router/src/reasoning_parser/traits.rs
@@ -69,6 +69,11 @@ pub trait ReasoningParser: Send + Sync {
 
     /// Get the model type this parser is designed for.
     fn model_type(&self) -> &str;
+
+    /// Check if the parser is currently in reasoning mode.
+    ///
+    /// Returns true if the parser is currently parsing reasoning content.
+    fn is_in_reasoning(&self) -> bool;
 }
 
 /// Error types for reasoning parsing operations.
diff --git a/sgl-router/src/routers/grpc/pd_router.rs b/sgl-router/src/routers/grpc/pd_router.rs
index da3c2d7f0..362e27fcc 100644
--- a/sgl-router/src/routers/grpc/pd_router.rs
+++ b/sgl-router/src/routers/grpc/pd_router.rs
@@ -922,27 +922,31 @@ impl GrpcPDRouter {
                     stream_buffer.push_str(&delta);
 
                     // Reasoning content handling
-                    if separate_reasoning {
-                        let (normal_text, reasoning_chunk) = router.process_reasoning_stream(
-                            &delta,
-                            index,
-                            &mut reasoning_parsers,
-                            &request_id,
-                            &model,
-                            created,
-                        );
+                    let in_reasoning = if separate_reasoning {
+                        let (normal_text, reasoning_chunk, in_reasoning) = router
+                            .process_reasoning_stream(
+                                &delta,
+                                index,
+                                &mut reasoning_parsers,
+                                &request_id,
+                                &model,
+                                created,
+                            );
                         if let Some(chunk) = reasoning_chunk {
                             tx.send(Ok(bytes::Bytes::from(Self::format_sse_chunk(&chunk))))
                                 .map_err(|_| "Failed to send reasoning chunk".to_string())?;
                         }
                         delta = normal_text;
-                    }
+                        in_reasoning
+                    } else {
+                        false
+                    };
 
                     // Tool call handling
                     let tool_choice_enabled =
                         !matches!(tool_choice, Some(ToolChoice::Value(ToolChoiceValue::None)));
 
-                    if tool_choice_enabled && tools.is_some() {
+                    if !in_reasoning && tool_choice_enabled && tools.is_some() {
                         let (should_skip, tool_chunks) = router
                             .process_tool_calls_stream(
                                 &delta,
@@ -1173,16 +1177,18 @@ impl GrpcPDRouter {
         request_id: &str,
         model: &str,
         created: u64,
-    ) -> (String, Option<ChatCompletionStreamResponse>) {
+    ) -> (String, Option<ChatCompletionStreamResponse>, bool) {
         // Get or create parser for this index
         reasoning_parsers
             .entry(index)
             .or_insert_with(|| self.reasoning_parser_factory.get_pooled(model));
 
         if let Some(pooled_parser) = reasoning_parsers.get(&index) {
-            let parse_result = {
+            let (parse_result, in_reasoning) = {
                 let mut parser = pooled_parser.lock().unwrap();
-                parser.parse_reasoning_streaming_incremental(delta)
+                let result = parser.parse_reasoning_streaming_incremental(delta);
+                let in_reasoning = parser.is_in_reasoning();
+                (result, in_reasoning)
             };
 
             match parse_result {
@@ -1214,7 +1220,7 @@ impl GrpcPDRouter {
                     } else {
                         None
                     };
-                    return (normal_text, chunk);
+                    return (normal_text, chunk, in_reasoning);
                 }
                 Err(e) => {
                     warn!("Reasoning parsing error: {}", e);
@@ -1222,7 +1228,7 @@ impl GrpcPDRouter {
             }
         }
 
-        (delta.to_string(), None)
+        (delta.to_string(), None, false)
     }
 
     /// Helper: Process tool calls in streaming mode
diff --git a/sgl-router/src/routers/grpc/router.rs b/sgl-router/src/routers/grpc/router.rs
index 06832c764..e90df51fd 100644
--- a/sgl-router/src/routers/grpc/router.rs
+++ b/sgl-router/src/routers/grpc/router.rs
@@ -494,16 +494,18 @@ impl GrpcRouter {
         request_id: &str,
         model: &str,
         created: u64,
-    ) -> (String, Option<ChatCompletionStreamResponse>) {
+    ) -> (String, Option<ChatCompletionStreamResponse>, bool) {
         // Get or create parser for this index
         reasoning_parsers
             .entry(index)
             .or_insert_with(|| self.reasoning_parser_factory.get_pooled(model));
 
         if let Some(pooled_parser) = reasoning_parsers.get(&index) {
-            let parse_result = {
+            let (parse_result, in_reasoning) = {
                 let mut parser = pooled_parser.lock().unwrap();
-                parser.parse_reasoning_streaming_incremental(delta)
+                let result = parser.parse_reasoning_streaming_incremental(delta);
+                let in_reasoning = parser.is_in_reasoning();
+                (result, in_reasoning)
             };
 
             match parse_result {
@@ -535,7 +537,7 @@ impl GrpcRouter {
                     } else {
                         None
                     };
-                    return (normal_text, chunk);
+                    return (normal_text, chunk, in_reasoning);
                 }
                 Err(e) => {
                     warn!("Reasoning parsing error: {}", e);
@@ -543,7 +545,7 @@ impl GrpcRouter {
             }
         }
 
-        (delta.to_string(), None)
+        (delta.to_string(), None, false)
     }
 
     /// Helper: Process tool calls in streaming mode
@@ -901,27 +903,31 @@ impl GrpcRouter {
                     stream_buffer.push_str(&delta);
 
                     // Reasoning content handling
-                    if separate_reasoning {
-                        let (normal_text, reasoning_chunk) = router.process_reasoning_stream(
-                            &delta,
-                            index,
-                            &mut reasoning_parsers,
-                            &request_id,
-                            &model,
-                            created,
-                        );
+                    let in_reasoning = if separate_reasoning {
+                        let (normal_text, reasoning_chunk, in_reasoning) = router
+                            .process_reasoning_stream(
+                                &delta,
+                                index,
+                                &mut reasoning_parsers,
+                                &request_id,
+                                &model,
+                                created,
+                            );
                         if let Some(chunk) = reasoning_chunk {
                             tx.send(Ok(Bytes::from(Self::format_sse_chunk(&chunk))))
                                 .map_err(|_| "Failed to send reasoning chunk".to_string())?;
                         }
                         delta = normal_text;
-                    }
+                        in_reasoning
+                    } else {
+                        false
+                    };
 
                     // Tool call handling
                     let tool_choice_enabled =
                         !matches!(tool_choice, Some(ToolChoice::Value(ToolChoiceValue::None)));
 
-                    if tool_choice_enabled && tools.is_some() {
+                    if !in_reasoning && tool_choice_enabled && tools.is_some() {
                         let (should_skip, tool_chunks) = router
                             .process_tool_calls_stream(
                                 &delta,