Skip to content

fix(anthropic): stop double-counting output tokens on streamed messages #4377

New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Open
arthi-arumugam-git wants to merge 3 commits into
base: main
Choose a base branch
from
Open

fix(anthropic): stop double-counting output tokens on streamed messages #4377

Show file tree
Hide file tree
Changes from all commits
Commits
Show all changes
3 commits
Select commit Hold shift + click to select a range
b3722db
fix(anthropic): stop double-counting output tokens on streamed messages
arthi-arumugam-git Jul 26, 2026
7c33857
Merge branch 'main' into fix/anthropic-streaming-output-token-double-…
arthi-arumugam-git Aug 13, 2026
36a043a
test(anthropic): drop one-token overcount in prompt-caching stream as…
arthi-arumugam-git Nov 5, 2026
File filter

Filter by extension

Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
13 changes: 6 additions & 7 deletions ...ntelemetry-instrumentation-anthropic/opentelemetry/instrumentation/anthropic/streaming.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -62,13 +62,12 @@ def _process_response_item(item, complete_response):
event["finish_reason"] = item.delta.stop_reason
if item.usage:
if "usage" in complete_response:
item_output_tokens = dict(item.usage).get("output_tokens", 0)
existing_output_tokens = complete_response["usage"].get(
"output_tokens", 0
)
complete_response["usage"]["output_tokens"] = (
item_output_tokens + existing_output_tokens
)
# message_delta carries the running total for the whole message,
# not an increment, so it replaces the partial count reported by
# message_start rather than adding to it.
item_output_tokens = dict(item.usage).get("output_tokens")
if item_output_tokens is not None:
complete_response["usage"]["output_tokens"] = item_output_tokens
else:
complete_response["usage"] = dict(item.usage)

Expand Down
1 change: 1 addition & 0 deletions packages/opentelemetry-instrumentation-anthropic/tests/test_messages.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -680,6 +680,7 @@ def test_anthropic_message_streaming_legacy(
assert text_parts[0]["content"] == response_content
assert output_messages[-1]["role"] == "assistant"
assert anthropic_span.attributes[GenAIAttributes.GEN_AI_USAGE_INPUT_TOKENS] == 17
assert anthropic_span.attributes[GenAIAttributes.GEN_AI_USAGE_OUTPUT_TOKENS] == 171
assert (
anthropic_span.attributes[GenAIAttributes.GEN_AI_USAGE_OUTPUT_TOKENS]
+ anthropic_span.attributes[GenAIAttributes.GEN_AI_USAGE_INPUT_TOKENS]
Expand Down
13 changes: 7 additions & 6 deletions packages/opentelemetry-instrumentation-anthropic/tests/test_prompt_caching.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -759,7 +759,7 @@ def test_anthropic_prompt_caching_stream_legacy(
assert json.loads(cache_creation_span.attributes[GenAIAttributes.GEN_AI_OUTPUT_MESSAGES])[-1]["role"] == "assistant"
assert json.loads(cache_read_span.attributes[GenAIAttributes.GEN_AI_OUTPUT_MESSAGES])[-1]["role"] == "assistant"

_verify_caching_attributes(cache_creation_span, cache_read_span, 1169, 202, 222, 1165)
_verify_caching_attributes(cache_creation_span, cache_read_span, 1169, 201, 221, 1165)

# verify metrics
metrics_data = reader.get_metrics_data()
Expand Down Expand Up @@ -820,7 +820,7 @@ def test_anthropic_prompt_caching_stream_with_events_with_content(
cache_creation_span = spans[0]
cache_read_span = spans[1]

_verify_caching_attributes(cache_creation_span, cache_read_span, 1169, 202, 222, 1165)
_verify_caching_attributes(cache_creation_span, cache_read_span, 1169, 201, 221, 1165)

# verify metrics
metrics_data = reader.get_metrics_data()
Expand Down Expand Up @@ -986,7 +986,7 @@ def test_anthropic_prompt_caching_stream_with_events_with_no_content(
cache_creation_span = spans[0]
cache_read_span = spans[1]

_verify_caching_attributes(cache_creation_span, cache_read_span, 1169, 202, 222, 1165)
_verify_caching_attributes(cache_creation_span, cache_read_span, 1169, 201, 221, 1165)

# verify metrics
metrics_data = reader.get_metrics_data()
Expand Down Expand Up @@ -1089,7 +1089,7 @@ async def test_anthropic_prompt_caching_async_stream_legacy(
assert cr_input[0]["role"] == "user"
assert text == cr_input[0]["parts"][0]["content"]

_verify_caching_attributes(cache_creation_span, cache_read_span, 1171, 290, 257, 1167)
_verify_caching_attributes(cache_creation_span, cache_read_span, 1171, 289, 256, 1167)

# verify metrics
metrics_data = reader.get_metrics_data()
Expand Down Expand Up @@ -1151,7 +1151,7 @@ async def test_anthropic_prompt_caching_async_stream_with_events_with_content(
cache_creation_span = spans[0]
cache_read_span = spans[1]

_verify_caching_attributes(cache_creation_span, cache_read_span, 1171, 290, 257, 1167)
_verify_caching_attributes(cache_creation_span, cache_read_span, 1171, 289, 256, 1167)

# verify metrics
metrics_data = reader.get_metrics_data()
Expand Down Expand Up @@ -1328,7 +1328,7 @@ async def test_anthropic_prompt_caching_async_stream_with_events_with_no_content
cache_creation_span = spans[0]
cache_read_span = spans[1]

_verify_caching_attributes(cache_creation_span, cache_read_span, 1171, 290, 257, 1167)
_verify_caching_attributes(cache_creation_span, cache_read_span, 1171, 289, 256, 1167)

# verify metrics
metrics_data = reader.get_metrics_data()
Expand Down Expand Up @@ -1372,3 +1372,4 @@ def assert_message_in_logs(log: ReadableLogRecord, event_name: str, expected_con
else:
assert log.log_record.body
assert dict(log.log_record.body) == expected_content