diff --git a/src/agents/items.py b/src/agents/items.py
index 991a7f877..76be5122b 100644
--- a/src/agents/items.py
+++ b/src/agents/items.py
@@ -356,6 +356,13 @@ class ModelResponse:
     be passed to `Runner.run`.
     """
 
+    logprobs: list[Any] | None = None
+    """Token log probabilities from the model response.
+    Only populated when using the chat completions API with `top_logprobs` set in ModelSettings.
+    Each element corresponds to a token and contains the token string, log probability, and
+    optionally the top alternative tokens with their log probabilities.
+    """
+
     def to_input_items(self) -> list[TResponseInputItem]:
         """Convert the output into a list of input items suitable for passing to the model."""
         # We happen to know that the shape of the Pydantic output items are the same as the
diff --git a/src/agents/models/openai_chatcompletions.py b/src/agents/models/openai_chatcompletions.py
index 594848d3e..ac696dbda 100644
--- a/src/agents/models/openai_chatcompletions.py
+++ b/src/agents/models/openai_chatcompletions.py
@@ -129,10 +129,15 @@ async def get_response(
 
             items = Converter.message_to_output_items(message) if message is not None else []
 
+            logprobs_data = None
+            if first_choice and first_choice.logprobs and first_choice.logprobs.content:
+                logprobs_data = [lp.model_dump() for lp in first_choice.logprobs.content]
+
             return ModelResponse(
                 output=items,
                 usage=usage,
                 response_id=None,
+                logprobs=logprobs_data,
             )
 
     async def stream_response(