From c08f218bf8929579b1eb69ef99246f66426ef5f7 Mon Sep 17 00:00:00 2001
From: Thiago Moretto <thiago.moretto@gmail.com>
Date: Wed, 7 Jan 2026 09:03:44 -0300
Subject: [PATCH] fix: account for thought tokens on gemini models

---
 lib/crewai/src/crewai/llms/base_llm.py                    | 4 ++++
 lib/crewai/src/crewai/llms/providers/gemini/completion.py | 1 +
 lib/crewai/tests/llms/google/test_google.py               | 3 +++
 3 files changed, 8 insertions(+)

diff --git a/lib/crewai/src/crewai/llms/base_llm.py b/lib/crewai/src/crewai/llms/base_llm.py
index c09c26453..8b9c9e241 100644
--- a/lib/crewai/src/crewai/llms/base_llm.py
+++ b/lib/crewai/src/crewai/llms/base_llm.py
@@ -569,6 +569,10 @@ class BaseLLM(ABC):
             or usage_data.get("input_tokens")
             or 0
         )
+        + (
+            usage_data.get("thoughts_token_count")
+            or 0
+        )
 
         completion_tokens = (
             usage_data.get("completion_tokens")
diff --git a/lib/crewai/src/crewai/llms/providers/gemini/completion.py b/lib/crewai/src/crewai/llms/providers/gemini/completion.py
index b268f07de..b6c893aa9 100644
--- a/lib/crewai/src/crewai/llms/providers/gemini/completion.py
+++ b/lib/crewai/src/crewai/llms/providers/gemini/completion.py
@@ -982,6 +982,7 @@ class GeminiCompletion(BaseLLM):
             usage = response.usage_metadata
             return {
                 "prompt_token_count": getattr(usage, "prompt_token_count", 0),
+                "thoughts_token_count": getattr(usage, "thoughts_token_count", 0),
                 "candidates_token_count": getattr(usage, "candidates_token_count", 0),
                 "total_token_count": getattr(usage, "total_token_count", 0),
                 "total_tokens": getattr(usage, "total_token_count", 0),
diff --git a/lib/crewai/tests/llms/google/test_google.py b/lib/crewai/tests/llms/google/test_google.py
index 37f591de6..d24f07a93 100644
--- a/lib/crewai/tests/llms/google/test_google.py
+++ b/lib/crewai/tests/llms/google/test_google.py
@@ -79,6 +79,7 @@ def test_gemini_tool_use_conversation_flow():
         mock_response.text = "Based on the weather data, it's a beautiful day in San Francisco with sunny skies and 75°F temperature."
         mock_response.usage_metadata = Mock()
         mock_response.usage_metadata.prompt_token_count = 100
+        mock_response.usage_metadata.thoughts_token_count = 50
         mock_response.usage_metadata.candidates_token_count = 50
         mock_response.usage_metadata.total_token_count = 150
 
@@ -630,6 +631,7 @@ def test_gemini_token_usage_tracking():
         mock_response.candidates = []
         mock_response.usage_metadata = MagicMock(
             prompt_token_count=50,
+            thoughts_token_count=25,
             candidates_token_count=25,
             total_token_count=75
         )
@@ -643,6 +645,7 @@ def test_gemini_token_usage_tracking():
         # Verify token usage was extracted
         usage = llm._extract_token_usage(mock_response)
         assert usage["prompt_token_count"] == 50
+        assert usage["thoughts_token_count"] == 25
         assert usage["candidates_token_count"] == 25
         assert usage["total_token_count"] == 75
         assert usage["total_tokens"] == 75