diff --git a/metagpt/utils/token_counter.py b/metagpt/utils/token_counter.py
index d68b9e752..e977884ab 100644
--- a/metagpt/utils/token_counter.py
+++ b/metagpt/utils/token_counter.py
@@ -403,20 +403,25 @@ SPARK_TOKENS = {
 }
 
 
+def count_claude_message_tokens(messages: list[dict], model: str) -> int:
+    # rough estimation for models newer than claude-2.1, needs api_key or auth_token
+    ac = anthropic.Client()
+    system_prompt = ""
+    new_messages = []
+    for msg in messages:
+        if msg.get("role") == "system":
+            system_prompt = msg.get("content")
+        else:
+            new_messages.append(msg)
+    num_tokens = ac.beta.messages.count_tokens(messages=new_messages, model=model, system=system_prompt)
+    return num_tokens.input_tokens
+
+
 def count_message_tokens(messages, model="gpt-3.5-turbo-0125"):
     """Return the number of tokens used by a list of messages."""
     if "claude" in model:
-        # rough estimation for models newer than claude-2.1, needs api_key or auth_token
-        ac = anthropic.Client()
-        system_prompt = ""
-        new_messages = []
-        for msg in messages:
-            if msg.get("role") == "system":
-                system_prompt = msg.get("content")
-            else:
-                new_messages.append(msg)
-        num_tokens = ac.beta.messages.count_tokens(messages=new_messages, model=model, system=system_prompt)
-        return num_tokens.get("input_tokens", 0)
+        num_tokens = count_claude_message_tokens(messages, model)
+        return num_tokens.input_tokens
     try:
         encoding = tiktoken.encoding_for_model(model)
     except KeyError:
@@ -504,8 +509,8 @@ def count_output_tokens(string: str, model: str) -> int:
         int: The number of tokens in the text string.
     """
     if "claude" in model:
-        vo = anthropic.Client()
-        num_tokens = vo.count_tokens(string)
+        messages = [{"role": "assistant", "content": string}]
+        num_tokens = count_claude_message_tokens(messages, model)
         return num_tokens
     try:
         encoding = tiktoken.encoding_for_model(model)