diff --git a/examples/filters/langfuse_filter_pipeline.py b/examples/filters/langfuse_filter_pipeline.py
index ada6d88..2bee9b6 100644
--- a/examples/filters/langfuse_filter_pipeline.py
+++ b/examples/filters/langfuse_filter_pipeline.py
@@ -9,15 +9,20 @@ requirements: langfuse
 """
 
 from typing import List, Optional
-from schemas import OpenAIChatMessage
 import os
 import uuid
 
-from utils.pipelines.main import get_last_user_message, get_last_assistant_message
+from utils.pipelines.main import get_last_assistant_message
 from pydantic import BaseModel
 from langfuse import Langfuse
 from langfuse.api.resources.commons.errors.unauthorized_error import UnauthorizedError
 
+def get_last_assistant_message_obj(messages: List[dict]) -> dict:
+    for message in reversed(messages):
+        if message["role"] == "assistant":
+            return message
+    return {}
+
 
 class Pipeline:
     class Valves(BaseModel):
@@ -109,21 +114,28 @@ class Pipeline:
 
     async def outlet(self, body: dict, user: Optional[dict] = None) -> dict:
         print(f"outlet:{__name__}")
+        print(f"Received body: {body}")
         if body["chat_id"] not in self.chat_generations:
             return body
 
         generation = self.chat_generations[body["chat_id"]]
         assistant_message = get_last_assistant_message(body["messages"])
 
-        # Extract usage information
-        info = assistant_message.get("info", {})
+        
+        # Extract usage information for models that support it
         usage = None
-        if "prompt_tokens" in info and "completion_tokens" in info:
-            usage = {
-                "input": info["prompt_tokens"],
-                "output": info["completion_tokens"],
-                "unit": "TOKENS",
-            }
+        assistant_message_obj = get_last_assistant_message_obj(body["messages"])
+        if assistant_message_obj:
+            info = assistant_message_obj.get("info", {})
+            if isinstance(info, dict):
+                prompt_eval_count = info.get("prompt_eval_count")
+                eval_count = info.get("eval_count")
+                if prompt_eval_count is not None and eval_count is not None:
+                    usage = {
+                        "input": prompt_eval_count,
+                        "output": eval_count,
+                        "unit": "TOKENS",
+                    }
 
         # Update generation
         generation.end(