fix(openai): report request attributes in responses API instrumentation (#3471)

Author: galkleinman

packages/opentelemetry-instrumentation-openai/opentelemetry/instrumentation/openai/v1/responses_wrappers.py

@@ -51,6 +51,8 @@
 from typing_extensions import NotRequired
 
 from opentelemetry.instrumentation.openai.shared import (
+    _extract_model_name_from_provider_format,
+    _set_request_attributes,
     _set_span_attribute,
     model_as_dict,
 )
@@ -189,12 +191,26 @@ def process_content_block(
 
 
 @dont_throw
+def prepare_kwargs_for_shared_attributes(kwargs):
+    """
+    Prepare kwargs for the shared _set_request_attributes function.
+    Maps responses API specific parameters to the common format.
+    """
+    prepared_kwargs = kwargs.copy()
+
+    # Map max_output_tokens to max_tokens for the shared function
+    if "max_output_tokens" in kwargs:
+        prepared_kwargs["max_tokens"] = kwargs["max_output_tokens"]
+
+    return prepared_kwargs
+
+
 def set_data_attributes(traced_response: TracedData, span: Span):
-    _set_span_attribute(span, GenAIAttributes.GEN_AI_SYSTEM, "openai")
-    _set_span_attribute(span, GenAIAttributes.GEN_AI_REQUEST_MODEL, traced_response.request_model)
     _set_span_attribute(span, GenAIAttributes.GEN_AI_RESPONSE_ID, traced_response.response_id)
-    _set_span_attribute(span, GenAIAttributes.GEN_AI_RESPONSE_MODEL, traced_response.response_model)
-    _set_span_attribute(span, OpenAIAttributes.OPENAI_REQUEST_SERVICE_TIER, traced_response.request_service_tier)
+
+    response_model = _extract_model_name_from_provider_format(traced_response.response_model)
+    _set_span_attribute(span, GenAIAttributes.GEN_AI_RESPONSE_MODEL, response_model)
+
     _set_span_attribute(span, OpenAIAttributes.OPENAI_RESPONSE_SERVICE_TIER, traced_response.response_service_tier)
     if usage := traced_response.usage:
         _set_span_attribute(span, GenAIAttributes.GEN_AI_USAGE_INPUT_TOKENS, usage.input_tokens)
@@ -445,6 +461,7 @@ def responses_get_or_create_wrapper(tracer: Tracer, wrapped, instance, args, kwa
                 kind=SpanKind.CLIENT,
                 start_time=start_time,
             )
+            _set_request_attributes(span, prepare_kwargs_for_shared_attributes(kwargs), instance)
 
             return ResponseStream(
                 span=span,
@@ -503,6 +520,7 @@ def responses_get_or_create_wrapper(tracer: Tracer, wrapped, instance, args, kwa
                 start_time if traced_data is None else int(traced_data.start_time)
             ),
         )
+        _set_request_attributes(span, prepare_kwargs_for_shared_attributes(kwargs), instance)
         span.set_attribute(ERROR_TYPE, e.__class__.__name__)
         span.record_exception(e)
         span.set_status(StatusCode.ERROR, str(e))
@@ -568,6 +586,7 @@ def responses_get_or_create_wrapper(tracer: Tracer, wrapped, instance, args, kwa
             kind=SpanKind.CLIENT,
             start_time=int(traced_data.start_time),
         )
+        _set_request_attributes(span, prepare_kwargs_for_shared_attributes(kwargs), instance)
         set_data_attributes(traced_data, span)
         span.end()
 
@@ -591,6 +610,7 @@ async def async_responses_get_or_create_wrapper(
                 kind=SpanKind.CLIENT,
                 start_time=start_time,
             )
+            _set_request_attributes(span, prepare_kwargs_for_shared_attributes(kwargs), instance)
 
             return ResponseStream(
                 span=span,
@@ -645,6 +665,7 @@ async def async_responses_get_or_create_wrapper(
                 start_time if traced_data is None else int(traced_data.start_time)
             ),
         )
+        _set_request_attributes(span, prepare_kwargs_for_shared_attributes(kwargs), instance)
         span.set_attribute(ERROR_TYPE, e.__class__.__name__)
         span.record_exception(e)
         span.set_status(StatusCode.ERROR, str(e))
@@ -711,6 +732,7 @@ async def async_responses_get_or_create_wrapper(
             kind=SpanKind.CLIENT,
             start_time=int(traced_data.start_time),
         )
+        _set_request_attributes(span, prepare_kwargs_for_shared_attributes(kwargs), instance)
         set_data_attributes(traced_data, span)
         span.end()
 
@@ -735,6 +757,7 @@ def responses_cancel_wrapper(tracer: Tracer, wrapped, instance, args, kwargs):
             start_time=existing_data.start_time,
             record_exception=True,
         )
+        _set_request_attributes(span, prepare_kwargs_for_shared_attributes(kwargs), instance)
         span.record_exception(Exception("Response cancelled"))
         set_data_attributes(existing_data, span)
         span.end()
@@ -761,6 +784,7 @@ async def async_responses_cancel_wrapper(
             start_time=existing_data.start_time,
             record_exception=True,
         )
+        _set_request_attributes(span, prepare_kwargs_for_shared_attributes(kwargs), instance)
         span.record_exception(Exception("Response cancelled"))
         set_data_attributes(existing_data, span)
         span.end()

packages/opentelemetry-instrumentation-openai/tests/traces/cassettes/test_responses/test_responses_with_request_params.yaml

@@ -0,0 +1,111 @@
+interactions:
+- request:
+    body: '{"input": "What is the capital of France?", "max_output_tokens": 100, "model":
+      "gpt-4.1-nano", "temperature": 0.7, "top_p": 0.9}'
+    headers:
+      accept:
+      - application/json
+      accept-encoding:
+      - gzip, deflate
+      connection:
+      - keep-alive
+      content-length:
+      - '128'
+      content-type:
+      - application/json
+      host:
+      - api.openai.com
+      user-agent:
+      - OpenAI/Python 1.99.7
+      x-stainless-arch:
+      - arm64
+      x-stainless-async:
+      - 'false'
+      x-stainless-lang:
+      - python
+      x-stainless-os:
+      - MacOS
+      x-stainless-package-version:
+      - 1.99.7
+      x-stainless-read-timeout:
+      - '600'
+      x-stainless-retry-count:
+      - '0'
+      x-stainless-runtime:
+      - CPython
+      x-stainless-runtime-version:
+      - 3.10.16
+    method: POST
+    uri: https://api.openai.com/v1/responses
+  response:
+    body:
+      string: !!binary |
+        H4sIAAAAAAAAA3RUwW7bMAy95ysEnZtCdp3YyQfsvMNuxWDQNp1qlUVBooIGRf59sBw78dbeLD7y
+        mXyP0udGCKk7eRTSY3C12pVlh6qBBiqV56jU/pDvurLEw65VVXbYVf2havt90eUNFi/7Tj6NFNT8
+        wZZnGrIBp3jrERi7GkYsK/eFqsp9WSUsMHAMY01LgzPIeCNroH0/eYp27KsHE3AKa2O0Pcmj+NwI
+        IYR0cEE/1nd4RkMOvdwIcU3J6D2NmI3GpIC281/qDhm0CWs0sI8ta7Kr+AAfNUV2kWumd0xgptSC
+        MZGpWzBrtoE6NGNjJ8fb4jnbWrC0zVW+26pimxU3zRKvPIrXNM401GLHEE7fulGpbJ9VyQ1V7opc
+        ZX3evFSwqxJzYuGLw8SDIcAJ78B3siewJcto7009NrainUXBD16qUwJYSwyzkK+/V6Chk/PUfIEk
+        oqOQv95QtOA0gxHUix8ebItCB/ETvA7Pcqm53r4WGunJpNYgBB0YLE/JY2JKkg48GINm7Rr7OO2X
+        83jWFEM9r3CdnFhcdZ4Gx3UL7RvW73j5FvM4aqjJPmZ4hEB2tb/Y9+T5IWl0Jw4D+Jl7WecAPfKl
+        1t1I3GtcrXZAf9Yt1qzn69BDNJMvMjB5fByTcXDogWMKq+fyFk363zrryQ9wPz/4nvImXW8dn9E3
+        FDRfpm3rdBzu13BS+o10O1kTmeQC3NdAMrn6YTnUEnSpx8N09tG2cBNWdjpAY+Y3I6YlXwbQdn1l
+        i6f/4w/vwDJmMrC7F6rVqP++BNVX8a9oF/O/Y2ZiMHcwzxcFY1ibPSBDBwwj/XVz/QsAAP//AwDZ
+        CKVDwQUAAA==
+    headers:
+      CF-RAY:
+      - 9a427a739fc97d95-TLV
+      Connection:
+      - keep-alive
+      Content-Encoding:
+      - gzip
+      Content-Type:
+      - application/json
+      Date:
+      - Tue, 25 Nov 2025 16:21:20 GMT
+      Server:
+      - cloudflare
+      Set-Cookie:
+      - __cf_bm=972AABf9T_oEadbDKJ76kKF9Af2tTgOaTdaX4cxL8kg-1764087680-1.0.1.1-3xsObn4FVX3Nmp7VNdniyOmOOdjB3JnvdcHUEpHKZqNX2Q6j1k9MbrU0lGYlUVLrVb2Ls0gopzNAe6FA5AiAIvLVTtFzLT2780nDD_KBqxM;
+        path=/; expires=Tue, 25-Nov-25 16:51:20 GMT; domain=.api.openai.com; HttpOnly;
+        Secure; SameSite=None
+      - _cfuvid=w6zJyOCR9tAsFojGun.7hYM9_l7GfWdALH5Vd04TmeQ-1764087680812-0.0.1.1-604800000;
+        path=/; domain=.api.openai.com; HttpOnly; Secure; SameSite=None
+      Strict-Transport-Security:
+      - max-age=31536000; includeSubDomains; preload
+      Transfer-Encoding:
+      - chunked
+      X-Content-Type-Options:
+      - nosniff
+      alt-svc:
+      - h3=":443"; ma=86400
+      cf-cache-status:
+      - DYNAMIC
+      openai-organization:
+      - traceloop
+      openai-processing-ms:
+      - '2149'
+      openai-project:
+      - proj_tzz1TbPPOXaf6j9tEkVUBIAa
+      openai-version:
+      - '2020-10-01'
+      x-envoy-upstream-service-time:
+      - '2151'
+      x-ratelimit-limit-requests:
+      - '30000'
+      x-ratelimit-limit-tokens:
+      - '150000000'
+      x-ratelimit-remaining-requests:
+      - '29999'
+      x-ratelimit-remaining-tokens:
+      - '149999967'
+      x-ratelimit-reset-requests:
+      - 2ms
+      x-ratelimit-reset-tokens:
+      - 0s
+      x-request-id:
+      - req_500c9ae7e13b4845be386000962eb7fd
+    status:
+      code: 200
+      message: OK
+version: 1

packages/opentelemetry-instrumentation-openai/tests/traces/test_responses.py

@@ -29,6 +29,31 @@ def test_responses(
     # assert span.attributes["gen_ai.prompt.0.role"] == "user"
 
 
+@pytest.mark.vcr
+def test_responses_with_request_params(
+    instrument_legacy, span_exporter: InMemorySpanExporter, openai_client: OpenAI
+):
+    """Test that request parameters like temperature, max_tokens, top_p are captured"""
+    _ = openai_client.responses.create(
+        model="gpt-4.1-nano",
+        input="What is the capital of France?",
+        temperature=0.7,
+        max_output_tokens=100,
+        top_p=0.9,
+    )
+    spans = span_exporter.get_finished_spans()
+    assert len(spans) == 1
+    span = spans[0]
+    assert span.name == "openai.response"
+    assert span.attributes["gen_ai.system"] == "openai"
+    assert span.attributes["gen_ai.request.model"] == "gpt-4.1-nano"
+
+    # Check that request parameters are captured
+    assert span.attributes["gen_ai.request.temperature"] == 0.7
+    assert span.attributes["gen_ai.request.max_tokens"] == 100
+    assert span.attributes["gen_ai.request.top_p"] == 0.9
+
+
 @pytest.mark.vcr
 def test_responses_with_service_tier(
     instrument_legacy, span_exporter: InMemorySpanExporter, openai_client: OpenAI