From 4657d43fb999972bc9c77f65a4ee5a40bb68b44c Mon Sep 17 00:00:00 2001
From: kerry <kerry@berri.ai>
Date: Wed, 16 Sep 2026 00:27:09 +0000
Subject: [PATCH] fix(anthropic): tolerate message_delta chunks without usage
 in streams

Co-Authored-By: Devin AI <158243242+devin-ai-integration[bot]@users.noreply.github.com>
---
 litellm/llms/anthropic/chat/handler.py            |  4 +++-
 litellm/types/llms/anthropic.py                   |  2 +-
 .../anthropic/chat/test_anthropic_chat_handler.py | 15 +++++++++++++++
 3 files changed, 19 insertions(+), 2 deletions(-)

diff --git a/litellm/llms/anthropic/chat/handler.py b/litellm/llms/anthropic/chat/handler.py
index 5d3ae444b42..4dd0deeb62b 100644
--- a/litellm/llms/anthropic/chat/handler.py
+++ b/litellm/llms/anthropic/chat/handler.py
@@ -1167,7 +1167,9 @@ class ModelResponseIterator:
         # (matches OpenAI behavior and non-streaming Anthropic implementation)
         if self.converted_response_format_tool:
             finish_reason = "stop"
-        usage: Final = self._handle_usage(anthropic_usage_chunk=message_delta["usage"])
+        usage: Final = (
+            self._handle_usage(anthropic_usage_chunk=message_delta["usage"]) if "usage" in message_delta else None
+        )
         container: Final = message_delta["delta"].get("container")
         return finish_reason, usage, container
 
diff --git a/litellm/types/llms/anthropic.py b/litellm/types/llms/anthropic.py
index d56ada07ed5..ecb8dbb2502 100644
--- a/litellm/types/llms/anthropic.py
+++ b/litellm/types/llms/anthropic.py
@@ -586,7 +586,7 @@ class MessageBlockDelta(TypedDict):
 
     type: Literal["message_delta"]
     delta: MessageDelta
-    usage: UsageDelta
+    usage: NotRequired[UsageDelta]
     context_management: NotRequired[ContextManagementResponse]
 
 
diff --git a/tests/test_litellm/llms/anthropic/chat/test_anthropic_chat_handler.py b/tests/test_litellm/llms/anthropic/chat/test_anthropic_chat_handler.py
index 83201aef143..ac05cdcf251 100644
--- a/tests/test_litellm/llms/anthropic/chat/test_anthropic_chat_handler.py
+++ b/tests/test_litellm/llms/anthropic/chat/test_anthropic_chat_handler.py
@@ -579,6 +579,21 @@ def test_text_only_streaming_has_index_zero():
             ), f"Expected index=0, got {parsed.choices[0].index}"
 
 
+def test_message_delta_without_usage_returns_chunk_with_no_usage():
+    """A message_delta event may carry no usage field; it must not raise."""
+    iterator = ModelResponseIterator(None, sync_stream=True)
+
+    model_response = iterator.chunk_parser(
+        {
+            "type": "message_delta",
+            "delta": {"stop_reason": "end_turn", "stop_sequence": None},
+        }
+    )
+
+    assert model_response.choices[0].finish_reason == "stop"
+    assert model_response.usage is None
+
+
 def test_streaming_thinking_deltas_count_reasoning_tokens_in_usage():
     """Anthropic streaming usage should account for emitted thinking deltas."""
     chunks = [