Skip to content

Open-source Python library for AI browser agents, plus Browser Use Cloud: stealth browsers and a hosted agent API.

fix: remove double-counting of reasoning_tokens in OpenAI usage

Chase Xu993a4c1f5b36
diff --git a/browser_use/llm/openai/chat.py b/browser_use/llm/openai/chat.py
index 8b75eb54c..390df689f 100644
--- a/browser_use/llm/openai/chat.py
+++ b/browser_use/llm/openai/chat.py
@@ -122,13 +122,9 @@ class ChatOpenAI(BaseChatModel):
def _get_usage(self, response: ChatCompletion) -> ChatInvokeUsage | None:
if response.usage is not None:
- completion_tokens = response.usage.completion_tokens
- completion_token_details = response.usage.completion_tokens_details
- if completion_token_details is not None:
- reasoning_tokens = completion_token_details.reasoning_tokens
- if reasoning_tokens is not None:
- completion_tokens += reasoning_tokens
-
+ # Note: completion_tokens already includes reasoning_tokens per OpenAI API docs.
+ # Unlike Google Gemini where thinking_tokens are reported separately,
+ # OpenAI's reasoning_tokens are a subset of completion_tokens.
usage = ChatInvokeUsage(
prompt_tokens=response.usage.prompt_tokens,
prompt_cached_tokens=response.usage.prompt_tokens_details.cached_tokens
@@ -137,7 +133,7 @@ class ChatOpenAI(BaseChatModel):
prompt_cache_creation_tokens=None,
prompt_image_tokens=None,
# Completion
- completion_tokens=completion_tokens,
+ completion_tokens=response.usage.completion_tokens,
total_tokens=response.usage.total_tokens,
)
else: