diff --git a/src/pipecat/services/anthropic/llm.py b/src/pipecat/services/anthropic/llm.py
index c028ec7aa..cd00f15b8 100644
--- a/src/pipecat/services/anthropic/llm.py
+++ b/src/pipecat/services/anthropic/llm.py
@@ -329,7 +329,7 @@ class AnthropicLLMService(LLMService[AnthropicLLMAdapter]):
 
     def _get_llm_invocation_params(self, context: LLMContext) -> AnthropicLLMInvocationParams:
         adapter = self.get_llm_adapter()
-        params: AnthropicLLMInvocationParams = adapter.get_llm_invocation_params(
+        params = adapter.get_llm_invocation_params(
             context,
             enable_prompt_caching=assert_given(self._settings.enable_prompt_caching),
             system_instruction=assert_given(self._settings.system_instruction),
diff --git a/src/pipecat/services/aws/llm.py b/src/pipecat/services/aws/llm.py
index 7266821a1..6e68c5c6d 100644
--- a/src/pipecat/services/aws/llm.py
+++ b/src/pipecat/services/aws/llm.py
@@ -283,7 +283,7 @@ class AWSBedrockLLMService(LLMService[AWSBedrockLLMAdapter]):
             self._settings.system_instruction
         )
         adapter = self.get_llm_adapter()
-        params: AWSBedrockLLMInvocationParams = adapter.get_llm_invocation_params(
+        params = adapter.get_llm_invocation_params(
             context, system_instruction=effective_instruction
         )
         messages = params["messages"]
@@ -372,7 +372,7 @@ class AWSBedrockLLMService(LLMService[AWSBedrockLLMAdapter]):
 
     def _get_llm_invocation_params(self, context: LLMContext) -> AWSBedrockLLMInvocationParams:
         adapter = self.get_llm_adapter()
-        params: AWSBedrockLLMInvocationParams = adapter.get_llm_invocation_params(
+        params = adapter.get_llm_invocation_params(
             context, system_instruction=assert_given(self._settings.system_instruction)
         )
         return params
diff --git a/src/pipecat/services/google/llm.py b/src/pipecat/services/google/llm.py
index 355291e1b..da9659872 100644
--- a/src/pipecat/services/google/llm.py
+++ b/src/pipecat/services/google/llm.py
@@ -21,7 +21,7 @@ from loguru import logger
 from PIL import Image
 from pydantic import BaseModel, Field
 
-from pipecat.adapters.services.gemini_adapter import GeminiLLMAdapter, GeminiLLMInvocationParams
+from pipecat.adapters.services.gemini_adapter import GeminiLLMAdapter
 from pipecat.frames.frames import (
     AssistantImageRawFrame,
     Frame,
@@ -292,7 +292,7 @@ class GoogleLLMService(LLMService[GeminiLLMAdapter]):
         tools = []
         effective_instruction = system_instruction or self._settings.system_instruction
         adapter = self.get_llm_adapter()
-        params: GeminiLLMInvocationParams = adapter.get_llm_invocation_params(
+        params = adapter.get_llm_invocation_params(
             context, system_instruction=effective_instruction
         )
         messages = params["messages"]
@@ -387,7 +387,7 @@ class GoogleLLMService(LLMService[GeminiLLMAdapter]):
 
     async def _stream_content(self, context: LLMContext) -> AsyncIterator[GenerateContentResponse]:
         adapter = self.get_llm_adapter()
-        params: GeminiLLMInvocationParams = adapter.get_llm_invocation_params(
+        params = adapter.get_llm_invocation_params(
             context, system_instruction=assert_given(self._settings.system_instruction)
         )
 
diff --git a/src/pipecat/services/openai/base_llm.py b/src/pipecat/services/openai/base_llm.py
index 144927a8c..72d4360a9 100644
--- a/src/pipecat/services/openai/base_llm.py
+++ b/src/pipecat/services/openai/base_llm.py
@@ -297,7 +297,7 @@ class BaseOpenAILLMService(LLMService[OpenAILLMAdapter]):
             f"{self}: Generating chat from context {adapter.get_messages_for_logging(context)}"
         )
 
-        params_from_context: OpenAILLMInvocationParams = adapter.get_llm_invocation_params(
+        params_from_context = adapter.get_llm_invocation_params(
             context,
             system_instruction=self._settings.system_instruction,
             convert_developer_to_user=not self.supports_developer_role,
@@ -374,7 +374,7 @@ class BaseOpenAILLMService(LLMService[OpenAILLMAdapter]):
         """
         effective_instruction = system_instruction or self._settings.system_instruction
         adapter = self.get_llm_adapter()
-        invocation_params: OpenAILLMInvocationParams = adapter.get_llm_invocation_params(
+        invocation_params = adapter.get_llm_invocation_params(
             context,
             system_instruction=effective_instruction,
             convert_developer_to_user=not self.supports_developer_role,
diff --git a/tests/test_llm_service.py b/tests/test_llm_service.py
index c7018d9f0..707c255f3 100644
--- a/tests/test_llm_service.py
+++ b/tests/test_llm_service.py
@@ -7,6 +7,8 @@
 import unittest
 from unittest.mock import AsyncMock, patch
 
+from pipecat.adapters.base_llm_adapter import BaseLLMAdapter
+from pipecat.adapters.services.open_ai_adapter import OpenAILLMAdapter
 from pipecat.frames.frames import (
     FunctionCallFromLLM,
     FunctionCallInProgressFrame,
@@ -39,6 +41,25 @@ class MockLLMService(LLMService):
         super().__init__(settings=settings, **kwargs)
 
 
+class TestUnparameterizedSubclass(unittest.TestCase):
+    """Backward-compat coverage: third-party providers subclass LLMService
+    without specifying a generic adapter parameter. That should keep working
+    after LLMService became `Generic[TAdapter]`.
+    """
+
+    def test_unparameterized_subclass_instantiates(self):
+        # MockLLMService is declared as `class MockLLMService(LLMService):`
+        # — no generic bracket. The TypeVar's `bound=BaseLLMAdapter` should
+        # resolve TAdapter to BaseLLMAdapter for callers that don't opt in.
+        service = MockLLMService()
+        adapter = service.get_llm_adapter()
+
+        # Default adapter_class is OpenAILLMAdapter; the runtime instance
+        # should reflect that, regardless of how generics are erased.
+        self.assertIsInstance(adapter, OpenAILLMAdapter)
+        self.assertIsInstance(adapter, BaseLLMAdapter)
+
+
 class TestLLMService(unittest.IsolatedAsyncioTestCase):
     async def _run_function_calls_inline(self, service: MockLLMService):
         async def run_inline(runner_items):