diff --git a/instrumentation/elastic-opentelemetry-instrumentation-openai/src/opentelemetry/instrumentation/openai/helpers.py b/instrumentation/elastic-opentelemetry-instrumentation-openai/src/opentelemetry/instrumentation/openai/helpers.py
index 55f1462..f33ca17 100644
--- a/instrumentation/elastic-opentelemetry-instrumentation-openai/src/opentelemetry/instrumentation/openai/helpers.py
+++ b/instrumentation/elastic-opentelemetry-instrumentation-openai/src/opentelemetry/instrumentation/openai/helpers.py
@@ -49,7 +49,7 @@
     # available since 1.29.0
     GEN_AI_REQUEST_ENCODING_FORMATS = "gen_ai.request.encoding_formats"
 
-from opentelemetry.metrics import Histogram
+from opentelemetry.metrics import Histogram, NoOpHistogram
 from opentelemetry.trace import Span
 from opentelemetry.util.types import Attributes
 
@@ -191,6 +191,9 @@ def _get_attributes_if_set(span: Span, names: Iterable) -> Attributes:
 
 
 def _record_token_usage_metrics(metric: Histogram, span: Span, usage: CompletionUsage):
+    if _is_metric_disabled(metric):
+        return  # Avoid reading back attributes for an unrecorded metric
+
     token_usage_metric_attrs = _get_attributes_if_set(
         span,
         (
@@ -208,7 +211,14 @@ def _record_token_usage_metrics(metric: Histogram, span: Span, usage: Completion
         metric.record(usage.completion_tokens, {**token_usage_metric_attrs, GEN_AI_TOKEN_TYPE: "output"})
 
 
+def _is_metric_disabled(metric: Histogram):
+    return isinstance(metric, NoOpHistogram)
+
+
 def _record_operation_duration_metric(metric: Histogram, span: Span, start: float):
+    if _is_metric_disabled(metric):
+        return  # Avoid reading back attributes for an unrecorded metric
+
     operation_duration_metric_attrs = _get_attributes_if_set(
         span,
         (
diff --git a/instrumentation/elastic-opentelemetry-instrumentation-openai/tests/conftest.py b/instrumentation/elastic-opentelemetry-instrumentation-openai/tests/conftest.py
index b4e7d8a..28eb218 100644
--- a/instrumentation/elastic-opentelemetry-instrumentation-openai/tests/conftest.py
+++ b/instrumentation/elastic-opentelemetry-instrumentation-openai/tests/conftest.py
@@ -16,6 +16,7 @@
 
 import json
 import os
+import re
 from typing import Sequence, Tuple, Union
 
 import openai
@@ -105,15 +106,12 @@ def instrument():
 @pytest.fixture
 def vcr_cassette_name(request):
     """
-    Strips `_async` from the test function name as they use the same data.
+    Strips `_async` and `_disabled` from test function names as they use the same data.
     """
     # Get the name of the test function
     test_name = request.node.name
 
-    # Remove '_async' from the test name
-    cassette_name = test_name.replace("_async", "")
-
-    return cassette_name
+    return re.sub(r"(_async|_disabled)", "", test_name)
 
 
 OPENAI_API_KEY = "test_openai_api_key"
diff --git a/instrumentation/elastic-opentelemetry-instrumentation-openai/tests/test_chat_completions.py b/instrumentation/elastic-opentelemetry-instrumentation-openai/tests/test_chat_completions.py
index e83a5c5..5194cd5 100644
--- a/instrumentation/elastic-opentelemetry-instrumentation-openai/tests/test_chat_completions.py
+++ b/instrumentation/elastic-opentelemetry-instrumentation-openai/tests/test_chat_completions.py
@@ -26,6 +26,7 @@
 from opentelemetry._events import Event
 from opentelemetry._logs import LogRecord
 from opentelemetry.instrumentation.openai import OpenAIInstrumentor
+from opentelemetry.metrics import NoOpMeterProvider
 from opentelemetry.semconv._incubating.attributes.gen_ai_attributes import (
     GEN_AI_OPENAI_REQUEST_RESPONSE_FORMAT,
     GEN_AI_OPENAI_REQUEST_SEED,
@@ -48,7 +49,7 @@
 )
 from opentelemetry.semconv.attributes.error_attributes import ERROR_TYPE
 from opentelemetry.semconv.attributes.server_attributes import SERVER_ADDRESS, SERVER_PORT
-from opentelemetry.trace import SpanKind, StatusCode
+from opentelemetry.trace import NoOpTracerProvider, SpanKind, StatusCode
 
 from .conftest import (
     address_and_port,
@@ -1010,6 +1011,34 @@ def test_chat_stream_with_include_usage_option(default_openai_env, trace_exporte
     )
 
 
+@pytest.mark.vcr()
+def test_chat_stream_with_include_usage_option_disabled(default_openai_env, instrument):
+    """
+    When OTEL_SDK_DISABLED=true, opentelemetry-instrument still instruments
+    OpenAI, just with NoOp tracer and meter providers. This ensures when NoOp,
+    instrumentation doesn't crash.
+    """
+
+    instrument.uninstrument()
+    instrumentor = OpenAIInstrumentor()
+    instrumentor.instrument(tracer_provider=NoOpTracerProvider(), meter_provider=NoOpMeterProvider())
+    try:
+        client = openai.OpenAI()
+
+        messages = [
+            {
+                "role": "user",
+                "content": TEST_CHAT_INPUT,
+            }
+        ]
+
+        chat_completion = client.chat.completions.create(model=TEST_CHAT_MODEL, messages=messages, stream=True)
+        chunks = [chunk.choices[0].delta.content or "" for chunk in chat_completion if chunk.choices]
+        assert "".join(chunks) == "South Atlantic Ocean."
+    finally:
+        instrumentor.uninstrument()
+
+
 @pytest.mark.skipif(OPENAI_VERSION < (1, 26, 0), reason="stream_options added in 1.26.0")
 @pytest.mark.integration
 def test_chat_stream_with_include_usage_option_and_capture_message_content_integration(