diff --git a/changelog/4018.fixed.md b/changelog/4018.fixed.md
new file mode 100644
index 0000000000..d7be87dcee
--- /dev/null
+++ b/changelog/4018.fixed.md
@@ -0,0 +1 @@
+- Fixed LLM tracing model attribution so `gen_ai.request.model` stays stable to the configured request model and `gen_ai.response.model` is populated from provider response data when it arrives, including OpenAI-compatible streaming integrations.
diff --git a/src/pipecat/services/openai/base_llm.py b/src/pipecat/services/openai/base_llm.py
index eb8ce3cc66..6472200033 100644
--- a/src/pipecat/services/openai/base_llm.py
+++ b/src/pipecat/services/openai/base_llm.py
@@ -45,6 +45,10 @@
 from pipecat.services.settings import NOT_GIVEN as _NOT_GIVEN
 from pipecat.services.settings import LLMSettings, _NotGiven
 from pipecat.utils.tracing.service_decorators import traced_llm
+from pipecat.utils.tracing.setup import is_tracing_available
+
+if is_tracing_available():
+    from opentelemetry import trace
 
 
 @dataclass
@@ -257,6 +261,11 @@ def set_full_model_name(self, full_model_name: str):
         """
         self._full_model_name = full_model_name
 
+        if is_tracing_available() and "trace" in globals():
+            current_span = trace.get_current_span()
+            if current_span.is_recording():
+                current_span.set_attribute("gen_ai.response.model", full_model_name)
+
     def get_full_model_name(self):
         """Get the current full model name.
 
diff --git a/src/pipecat/services/sambanova/llm.py b/src/pipecat/services/sambanova/llm.py
index 710a22db21..7446fc1f3f 100644
--- a/src/pipecat/services/sambanova/llm.py
+++ b/src/pipecat/services/sambanova/llm.py
@@ -188,6 +188,9 @@ async def _process_context(
                     )
                     await self.start_llm_usage_metrics(tokens)
 
+                if chunk.model and self.get_full_model_name() != chunk.model:
+                    self.set_full_model_name(chunk.model)
+
                 if chunk.choices is None or len(chunk.choices) == 0:
                     continue
 
diff --git a/src/pipecat/utils/tracing/service_attributes.py b/src/pipecat/utils/tracing/service_attributes.py
index 21a22341f2..22198bd509 100644
--- a/src/pipecat/utils/tracing/service_attributes.py
+++ b/src/pipecat/utils/tracing/service_attributes.py
@@ -187,6 +187,7 @@ def add_llm_span_attributes(
     span: "Span",
     service_name: str,
     model: str,
+    response_model: Optional[str] = None,
     stream: bool = True,
     messages: Optional[str] = None,
     output: Optional[str] = None,
@@ -204,7 +205,8 @@ def add_llm_span_attributes(
     Args:
         span: The span to add attributes to.
         service_name: Name of the LLM service (e.g., "openai").
-        model: Model name/identifier.
+        model: Requested model name/identifier.
+        response_model: Actual model name returned by the provider.
         stream: Whether streaming is enabled.
         messages: JSON-serialized messages.
         output: Aggregated output text from the LLM.
@@ -220,6 +222,8 @@ def add_llm_span_attributes(
     # Add standard attributes
     span.set_attribute("gen_ai.system", _get_gen_ai_system_from_service_name(service_name))
     span.set_attribute("gen_ai.request.model", model)
+    if response_model:
+        span.set_attribute("gen_ai.response.model", response_model)
     span.set_attribute("gen_ai.operation.name", "chat")
     span.set_attribute("gen_ai.output.type", "text")
     span.set_attribute("stream", stream)
diff --git a/src/pipecat/utils/tracing/service_decorators.py b/src/pipecat/utils/tracing/service_decorators.py
index f4764d2485..b561aff51c 100644
--- a/src/pipecat/utils/tracing/service_decorators.py
+++ b/src/pipecat/utils/tracing/service_decorators.py
@@ -52,7 +52,6 @@ def _get_model_name(service) -> str:
     """
     return (
         getattr(getattr(service, "_settings", None), "model", None)
-        or getattr(service, "_full_model_name", None)
         or getattr(service, "model_name", None)
         or getattr(service, "_model_name", None)
         or "unknown"
@@ -525,6 +524,10 @@ async def wrapped_start_llm_usage_metrics(tokens):
                                 "parameters": params,
                             }
 
+                            full_model_name = getattr(self, "_full_model_name", None)
+                            if full_model_name:
+                                attribute_kwargs["response_model"] = full_model_name
+
                             # Add optional attributes only if they exist
                             if serialized_messages:
                                 attribute_kwargs["messages"] = serialized_messages
diff --git a/tests/test_llm_tracing_models.py b/tests/test_llm_tracing_models.py
new file mode 100644
index 0000000000..fa19f3b7ab
--- /dev/null
+++ b/tests/test_llm_tracing_models.py
@@ -0,0 +1,90 @@
+#
+# Copyright (c) 2024-2026, Daily
+#
+# SPDX-License-Identifier: BSD 2-Clause License
+#
+
+"""Unit tests for LLM tracing model attributes."""
+
+import importlib.util
+from pathlib import Path
+from types import SimpleNamespace
+from typing import cast
+
+from pipecat.utils.tracing.service_attributes import add_llm_span_attributes
+from pipecat.utils.tracing.service_decorators import _get_model_name
+
+
+class _FakeSpan:
+    def __init__(self, recording: bool = True):
+        self.attributes = {}
+        self._recording = recording
+
+    def set_attribute(self, key, value):
+        self.attributes[key] = value
+
+    def is_recording(self):
+        return self._recording
+
+
+class _ServiceWithModels:
+    def __init__(self):
+        self._settings = SimpleNamespace(model="requested-model")
+        self._full_model_name = "response-model"
+
+
+def _load_base_llm_module():
+    module_path = Path(__file__).resolve().parents[1] / "src/pipecat/services/openai/base_llm.py"
+    spec = importlib.util.spec_from_file_location("test_base_llm_module", module_path)
+    assert spec is not None and spec.loader is not None
+    module = importlib.util.module_from_spec(spec)
+    spec.loader.exec_module(module)
+    return module
+
+
+def test_get_model_name_prefers_requested_model_over_response_model():
+    """Request model should not drift to a previously observed response model."""
+    service = _ServiceWithModels()
+
+    model_name = _get_model_name(service)
+
+    assert model_name == "requested-model"
+
+
+def test_add_llm_span_attributes_sets_response_model_when_provided():
+    """LLM span attributes should include request and response model names."""
+    span = _FakeSpan()
+
+    add_llm_span_attributes(
+        span=span,
+        service_name="OpenAILLMService",
+        model="requested-model",
+        response_model="resolved-model-2026-01-01",
+    )
+
+    assert span.attributes["gen_ai.request.model"] == "requested-model"
+    assert span.attributes["gen_ai.response.model"] == "resolved-model-2026-01-01"
+
+
+def test_set_full_model_name_updates_active_span_response_model():
+    """Response model should be written when response data is received."""
+    base_llm = _load_base_llm_module()
+    fake_span = _FakeSpan(recording=True)
+    fake_trace = SimpleNamespace(get_current_span=lambda: fake_span)
+    fake_service = SimpleNamespace(_full_model_name="")
+
+    original_is_tracing_available = base_llm.is_tracing_available
+    original_trace = getattr(base_llm, "trace", None)
+    try:
+        setattr(base_llm, "is_tracing_available", lambda: True)
+        setattr(base_llm, "trace", fake_trace)
+        base_llm.BaseOpenAILLMService.set_full_model_name(fake_service, "gpt-4o-2026-02-14")
+    finally:
+        setattr(base_llm, "is_tracing_available", original_is_tracing_available)
+        if original_trace is None:
+            delattr(base_llm, "trace")
+        else:
+            setattr(base_llm, "trace", cast(object, original_trace))
+
+    assert fake_service._full_model_name == "gpt-4o-2026-02-14"
+    assert fake_span.attributes["gen_ai.response.model"] == "gpt-4o-2026-02-14"