Update LLMs

src/pipecat/services/aws/llm.py

@@ -75,10 +75,12 @@ class AWSBedrockLLMSettings(LLMSettings):
     """Settings for AWS Bedrock LLM services.
 
     Parameters:
+        stop_sequences: List of strings that stop generation.
         latency: Performance mode - "standard" or "optimized".
         additional_model_request_fields: Additional model-specific parameters.
     """
 
+    stop_sequences: List[str] | None | _NotGiven = field(default_factory=lambda: NOT_GIVEN)
     latency: str | _NotGiven = field(default_factory=lambda: NOT_GIVEN)
     additional_model_request_fields: Dict[str, Any] | _NotGiven = field(
         default_factory=lambda: NOT_GIVEN
@@ -810,6 +812,10 @@ class AWSBedrockLLMService(LLMService):
                 deprecated parameters and *settings* are provided, *settings*
                 values take precedence.
             stop_sequences: List of strings that stop generation.
+
+                .. deprecated:: 0.0.105
+                    Use ``settings=AWSBedrockLLMSettings(stop_sequences=...)`` instead.
+
             client_config: Custom boto3 client configuration.
             retry_timeout_secs: Request timeout in seconds for retry logic.
             retry_on_timeout: Whether to retry the request once if it times out.
@@ -828,6 +834,7 @@ class AWSBedrockLLMService(LLMService):
             seed=None,
             filter_incomplete_user_turns=False,
             user_turn_completion_config=None,
+            stop_sequences=None,
             latency=None,
             additional_model_request_fields={},
         )
@@ -836,6 +843,9 @@ class AWSBedrockLLMService(LLMService):
         if model is not None:
             _warn_deprecated_param("model", AWSBedrockLLMSettings, "model")
             default_settings.model = model
+        if stop_sequences is not None:
+            _warn_deprecated_param("stop_sequences", AWSBedrockLLMSettings, "stop_sequences")
+            default_settings.stop_sequences = stop_sequences
 
         # 3. Apply params overrides — only if settings not provided
         if params is not None:
@@ -844,6 +854,8 @@ class AWSBedrockLLMService(LLMService):
                 default_settings.max_tokens = params.max_tokens
                 default_settings.temperature = params.temperature
                 default_settings.top_p = params.top_p
+                if params.stop_sequences:
+                    default_settings.stop_sequences = params.stop_sequences
                 default_settings.latency = params.latency
                 if isinstance(params.additional_model_request_fields, dict):
                     default_settings.additional_model_request_fields = (
@@ -856,14 +868,6 @@ class AWSBedrockLLMService(LLMService):
 
         super().__init__(settings=default_settings, **kwargs)
 
-        # Handle stop_sequences (not a settings field)
-        if stop_sequences is not None:
-            self._stop_sequences = stop_sequences
-        elif params is not None:
-            self._stop_sequences = params.stop_sequences or []
-        else:
-            self._stop_sequences = []
-
         # Initialize the AWS Bedrock client
         if not client_config:
             client_config = Config(
@@ -915,6 +919,8 @@ class AWSBedrockLLMService(LLMService):
             inference_config["temperature"] = self._settings.temperature
         if self._settings.top_p is not None:
             inference_config["topP"] = self._settings.top_p
+        if self._settings.stop_sequences:
+            inference_config["stopSequences"] = self._settings.stop_sequences
         return inference_config
 
     async def run_inference(

src/pipecat/services/aws/nova_sonic/llm.py

@@ -191,12 +191,12 @@ class AWSNovaSonicLLMSettings(LLMSettings):
     """Settings for AWS Nova Sonic LLM service.
 
     Parameters:
-        voice_id: Voice for speech synthesis.
+        voice: Voice identifier for speech synthesis.
         endpointing_sensitivity: Controls how quickly Nova Sonic decides the
             user has stopped speaking. Can be "LOW", "MEDIUM", or "HIGH".
     """
 
-    voice_id: str | _NotGiven = field(default_factory=lambda: NOT_GIVEN)
+    voice: str | _NotGiven = field(default_factory=lambda: NOT_GIVEN)
     endpointing_sensitivity: str | None | _NotGiven = field(default_factory=lambda: NOT_GIVEN)
 
 
@@ -250,7 +250,7 @@ class AWSNovaSonicLLMService(LLMService):
                 - Nova Sonic (the older model): see https://docs.aws.amazon.com/nova/latest/userguide/available-voices.html.
 
                 .. deprecated::
-                    Use ``settings=AWSNovaSonicLLMSettings(voice_id=...)`` instead.
+                    Use ``settings=AWSNovaSonicLLMSettings(voice=...)`` instead.
 
             params: Model parameters for audio configuration and inference.
 
@@ -273,7 +273,7 @@ class AWSNovaSonicLLMService(LLMService):
         # 1. Initialize default_settings with hardcoded defaults
         default_settings = AWSNovaSonicLLMSettings(
             model="amazon.nova-2-sonic-v1:0",
-            voice_id="matthew",
+            voice="matthew",
             temperature=0.7,
             max_tokens=1024,
             top_p=0.9,
@@ -291,8 +291,8 @@ class AWSNovaSonicLLMService(LLMService):
             _warn_deprecated_param("model", AWSNovaSonicLLMSettings, "model")
             default_settings.model = model
         if voice_id != "matthew":
-            _warn_deprecated_param("voice_id", AWSNovaSonicLLMSettings, "voice_id")
+            _warn_deprecated_param("voice_id", AWSNovaSonicLLMSettings, "voice")
-            default_settings.voice_id = voice_id
+            default_settings.voice = voice_id
 
         # 3. Apply params overrides — only if settings not provided
         if params is not None:
@@ -811,7 +811,7 @@ class AWSNovaSonicLLMService(LLMService):
                 "sampleRateHertz": {self._output_sample_rate},
                 "sampleSizeBits": {self._output_sample_size},
                 "channelCount": {self._output_channel_count},
-                "voiceId": "{self._settings.voice_id}",
+                "voiceId": "{self._settings.voice}",
                 "encoding": "base64",
                 "audioType": "SPEECH"
               }}{tools_config}

src/pipecat/services/azure/realtime/llm.py

@@ -35,6 +35,8 @@ class AzureRealtimeLLMService(OpenAIRealtimeLLMService):
     real-time audio and text communication capabilities as the base OpenAI service.
     """
 
+    _settings: AzureRealtimeLLMSettings
+
     def __init__(
         self,
         *,

src/pipecat/services/cerebras/llm.py

@@ -31,6 +31,8 @@ class CerebrasLLMService(OpenAILLMService):
     maintaining full compatibility with OpenAI's interface and functionality.
     """
 
+    _settings: CerebrasLLMSettings
+
     def __init__(
         self,
         *,

src/pipecat/services/deepseek/llm.py

@@ -31,6 +31,8 @@ class DeepSeekLLMService(OpenAILLMService):
     maintaining full compatibility with OpenAI's interface and functionality.
     """
 
+    _settings: DeepSeekLLMSettings
+
     def __init__(
         self,
         *,

src/pipecat/services/fireworks/llm.py

@@ -31,6 +31,8 @@ class FireworksLLMService(OpenAILLMService):
     maintaining full compatibility with OpenAI's interface and functionality.
     """
 
+    _settings: FireworksLLMSettings
+
     def __init__(
         self,
         *,

src/pipecat/services/google/gemini_live/llm.py

@@ -610,6 +610,7 @@ class GeminiLiveLLMSettings(LLMSettings):
     """Settings for Gemini Live LLM services.
 
     Parameters:
+        voice: TTS voice identifier (e.g. ``"Charon"``).
         modalities: Response modalities.
         language: Language for generation.
         media_resolution: Media resolution setting.
@@ -620,6 +621,7 @@ class GeminiLiveLLMSettings(LLMSettings):
         proactivity: Proactivity configuration.
     """
 
+    voice: str | _NotGiven = field(default_factory=lambda: NOT_GIVEN)
     modalities: GeminiModalities | _NotGiven = field(default_factory=lambda: NOT_GIVEN)
     language: Language | str | _NotGiven = field(default_factory=lambda: NOT_GIVEN)
     media_resolution: GeminiMediaResolution | _NotGiven = field(default_factory=lambda: NOT_GIVEN)
@@ -680,6 +682,9 @@ class GeminiLiveLLMService(LLMService):
                     Use ``settings=GeminiLiveLLMSettings(model=...)`` instead.
 
             voice_id: TTS voice identifier. Defaults to "Charon".
+
+                .. deprecated:: 0.0.105
+                    Use ``settings=GeminiLiveLLMSettings(voice=...)`` instead.
             start_audio_paused: Whether to start with audio input paused. Defaults to False.
             start_video_paused: Whether to start with video input paused. Defaults to False.
             system_instruction: System prompt for the model. Defaults to None.
@@ -712,6 +717,7 @@ class GeminiLiveLLMService(LLMService):
         # 1. Initialize default_settings with hardcoded defaults
         default_settings = GeminiLiveLLMSettings(
             model="models/gemini-2.5-flash-native-audio-preview-12-2025",
+            voice="Charon",
             frequency_penalty=None,
             max_tokens=4096,
             presence_penalty=None,
@@ -736,6 +742,9 @@ class GeminiLiveLLMService(LLMService):
         if model is not None:
             _warn_deprecated_param("model", GeminiLiveLLMSettings, "model")
             default_settings.model = model
+        if voice_id != "Charon":
+            _warn_deprecated_param("voice_id", GeminiLiveLLMSettings, "voice")
+            default_settings.voice = voice_id
 
         # 3. Apply params overrides — only if settings not provided
         if params is not None:
@@ -776,7 +785,6 @@ class GeminiLiveLLMService(LLMService):
 
         self._last_sent_time = 0
         self._base_url = base_url
-        self._voice_id = voice_id
         self._language_code = params.language if params is not None else Language.EN_US
 
         self._system_instruction_from_init = system_instruction
@@ -1184,7 +1192,7 @@ class GeminiLiveLLMService(LLMService):
                     response_modalities=[Modality(self._settings.modalities.value)],
                     speech_config=SpeechConfig(
                         voice_config=VoiceConfig(
-                            prebuilt_voice_config={"voice_name": self._voice_id}
+                            prebuilt_voice_config={"voice_name": self._settings.voice}
                         ),
                         language_code=self._settings.language,
                     ),

src/pipecat/services/google/gemini_live/llm_vertex.py

@@ -57,6 +57,8 @@ class GeminiLiveVertexLLMService(GeminiLiveLLMService):
     responses, and tool usage.
     """
 
+    _settings: GeminiLiveVertexLLMSettings
+
     def __init__(
         self,
         *,
@@ -90,6 +92,9 @@ class GeminiLiveVertexLLMService(GeminiLiveLLMService):
                     Use ``settings=GeminiLiveLLMSettings(model=...)`` instead.
 
             voice_id: TTS voice identifier. Defaults to "Charon".
+
+                .. deprecated:: 0.0.105
+                    Use ``settings=GeminiLiveVertexLLMSettings(voice=...)`` instead.
             start_audio_paused: Whether to start with audio input paused. Defaults to False.
             start_video_paused: Whether to start with video input paused. Defaults to False.
             system_instruction: System prompt for the model. Defaults to None.
@@ -132,6 +137,7 @@ class GeminiLiveVertexLLMService(GeminiLiveLLMService):
         # 1. Initialize default_settings with hardcoded defaults
         default_settings = GeminiLiveVertexLLMSettings(
             model="google/gemini-live-2.5-flash-native-audio",
+            voice="Charon",
             frequency_penalty=None,
             max_tokens=4096,
             presence_penalty=None,
@@ -156,6 +162,9 @@ class GeminiLiveVertexLLMService(GeminiLiveLLMService):
         if model is not None:
             _warn_deprecated_param("model", GeminiLiveVertexLLMSettings, "model")
             default_settings.model = model
+        if voice_id != "Charon":
+            _warn_deprecated_param("voice_id", GeminiLiveVertexLLMSettings, "voice")
+            default_settings.voice = voice_id
 
         # 3. Apply params overrides — only if settings not provided
         if params is not None:
@@ -193,7 +202,6 @@ class GeminiLiveVertexLLMService(GeminiLiveLLMService):
             # api_key is required by parent class, but actually not used with
             # Vertex
             api_key="dummy",
-            voice_id=voice_id,
             start_audio_paused=start_audio_paused,
             start_video_paused=start_video_paused,
             system_instruction=system_instruction,

src/pipecat/services/google/llm_openai.py

@@ -58,6 +58,8 @@ class GoogleLLMOpenAIBetaService(OpenAILLMService):
         https://ai.google.dev/gemini-api/docs/openai
     """
 
+    _settings: GoogleOpenAILLMSettings
+
     def __init__(
         self,
         *,

src/pipecat/services/google/llm_vertex.py

@@ -59,6 +59,8 @@ class GoogleVertexLLMService(GoogleLLMService):
         https://cloud.google.com/vertex-ai/generative-ai/docs/model-reference/inference
     """
 
+    _settings: GoogleVertexLLMSettings
+
     class InputParams(GoogleLLMService.InputParams):
         """Input parameters specific to Vertex AI.
 

src/pipecat/services/grok/llm.py

@@ -86,6 +86,8 @@ class GrokLLMService(OpenAILLMService):
     processing and reports final totals.
     """
 
+    _settings: GrokLLMSettings
+
     def __init__(
         self,
         *,

src/pipecat/services/grok/realtime/llm.py

@@ -13,7 +13,7 @@ https://docs.x.ai/docs/guides/voice/agent
 import base64
 import json
 import time
-from dataclasses import dataclass, field
+from dataclasses import dataclass
 from typing import Any, Optional
 
 from loguru import logger
@@ -56,7 +56,7 @@ from pipecat.processors.aggregators.llm_response_universal import (
 from pipecat.processors.aggregators.openai_llm_context import OpenAILLMContext
 from pipecat.processors.frame_processor import FrameDirection
 from pipecat.services.llm_service import FunctionCallFromLLM, LLMService
-from pipecat.services.settings import NOT_GIVEN, LLMSettings, _NotGiven, _warn_deprecated_param
+from pipecat.services.settings import LLMSettings
 from pipecat.utils.time import time_now_iso8601
 
 from . import events
@@ -88,15 +88,9 @@ class CurrentAudioResponse:
 
 @dataclass
 class GrokRealtimeLLMSettings(LLMSettings):
-    """Settings for Grok Realtime LLM services.
+    """Settings for Grok Realtime LLM services."""
 
-    Parameters:
+    pass
-        session_properties: Grok Realtime session configuration.
-    """
-
-    session_properties: events.SessionProperties | _NotGiven = field(
-        default_factory=lambda: NOT_GIVEN
-    )
 
 
 class GrokRealtimeLLMService(LLMService):
@@ -137,19 +131,13 @@ class GrokRealtimeLLMService(LLMService):
             base_url: WebSocket base URL for the realtime API.
                 Defaults to "wss://api.x.ai/v1/realtime".
             session_properties: Configuration properties for the realtime session.
-
-                .. deprecated::
-                    Use ``settings=GrokRealtimeLLMSettings(session_properties=...)``
-                    instead.
-
                 If None, uses default SessionProperties with voice "Ara".
                 To set a different voice, configure it in session_properties:
 
                     session_properties = events.SessionProperties(voice="Rex")
 
                 Available voices: Ara, Rex, Sal, Eve, Leo.
-            settings: Grok Realtime LLM settings. If provided together with deprecated
+            settings: Runtime-updatable settings for this service.
-                top-level parameters, the ``settings`` values take precedence.
             start_audio_paused: Whether to start with audio input paused. Defaults to False.
             **kwargs: Additional arguments passed to parent LLMService.
         """
@@ -165,17 +153,9 @@ class GrokRealtimeLLMService(LLMService):
             seed=None,
             filter_incomplete_user_turns=False,
             user_turn_completion_config=None,
-            session_properties=events.SessionProperties(),
         )
 
-        # 2. Apply direct init arg overrides (deprecated)
+        # 2. Apply settings delta (canonical API, always wins)
-        if session_properties is not None:
-            _warn_deprecated_param(
-                "session_properties", GrokRealtimeLLMSettings, "session_properties"
-            )
-            default_settings.session_properties = session_properties
-
-        # 4. Apply settings delta (canonical API, always wins)
         if settings is not None:
             default_settings.apply_update(settings)
 
@@ -187,6 +167,7 @@ class GrokRealtimeLLMService(LLMService):
 
         self.api_key = api_key
         self.base_url = base_url
+        self._session_properties = session_properties or events.SessionProperties()
 
         self._audio_input_paused = start_audio_paused
         self._websocket = None
@@ -235,13 +216,13 @@ class GrokRealtimeLLMService(LLMService):
             Configured sample rate or None if not manually configured.
             For PCMU/PCMA formats, returns 8000 Hz (G.711 standard).
         """
-        if not self._settings.session_properties.audio:
+        if not self._session_properties.audio:
             return None
 
         audio_config = (
-            self._settings.session_properties.audio.input
+            self._session_properties.audio.input
             if direction == "input"
-            else self._settings.session_properties.audio.output
+            else self._session_properties.audio.output
         )
 
         if audio_config and audio_config.format:
@@ -271,8 +252,8 @@ class GrokRealtimeLLMService(LLMService):
 
     def _is_turn_detection_enabled(self) -> bool:
         """Check if server-side VAD is enabled."""
-        if self._settings.session_properties.turn_detection:
+        if self._session_properties.turn_detection:
-            return self._settings.session_properties.turn_detection.type == "server_vad"
+            return self._session_properties.turn_detection.type == "server_vad"
         return False
 
     async def _handle_interruption(self):
@@ -339,7 +320,7 @@ class GrokRealtimeLLMService(LLMService):
             input_sample_rate: Sample rate for audio input (Hz).
             output_sample_rate: Sample rate for audio output (Hz).
         """
-        props = self._settings.session_properties
+        props = self._session_properties
         if not props.audio:
             props.audio = events.AudioConfiguration()
         if not props.audio.input:
@@ -395,7 +376,12 @@ class GrokRealtimeLLMService(LLMService):
         # directly. The frame.delta path falls through to super, which calls
         # _update_settings → our override handles the rest.
         if isinstance(frame, LLMUpdateSettingsFrame) and frame.delta is None:
-            self._settings.session_properties = events.SessionProperties(**frame.settings)
+            # Capture current audio config before replacing session properties.
+            input_rate = self._get_configured_sample_rate("input")
+            output_rate = self._get_configured_sample_rate("output")
+            self._session_properties = events.SessionProperties(**frame.settings)
+            if input_rate and output_rate:
+                self._ensure_audio_config(input_rate, output_rate)
             await self._send_session_update()
             await self.push_frame(frame, direction)
             return
@@ -498,29 +484,14 @@ class GrokRealtimeLLMService(LLMService):
             await self.push_error(error_msg=f"Error sending client event: {e}", exception=e)
 
     async def _update_settings(self, delta):
-        """Apply a settings delta, sending a session update if needed."""
+        """Apply a settings delta."""
-        # Capture current sample rates before the update replaces them.
-        input_rate = self._get_configured_sample_rate("input")
-        output_rate = self._get_configured_sample_rate("output")
-
         changed = await super()._update_settings(delta)
-
+        self._warn_unhandled_updated_settings(changed.keys())
-        if "session_properties" in changed:
-            if input_rate and output_rate:
-                self._ensure_audio_config(input_rate, output_rate)
-            else:
-                logger.warning(
-                    "Attempting to apply session properties update without configured sample rates. "
-                    "Audio configuration may be incomplete."
-                )
-            await self._send_session_update()
-
-        self._warn_unhandled_updated_settings(changed.keys() - {"session_properties"})
         return changed
 
     async def _send_session_update(self):
         """Update session settings on the server."""
-        settings = self._settings.session_properties
+        settings = self._session_properties
         adapter: GrokRealtimeLLMAdapter = self.get_llm_adapter()
 
         if self._context:

src/pipecat/services/groq/llm.py

@@ -30,6 +30,8 @@ class GroqLLMService(OpenAILLMService):
     maintaining full compatibility with OpenAI's interface and functionality.
     """
 
+    _settings: GroqLLMSettings
+
     def __init__(
         self,
         *,

src/pipecat/services/mistral/llm.py

@@ -33,6 +33,8 @@ class MistralLLMService(OpenAILLMService):
     maintaining full compatibility with OpenAI's interface and functionality.
     """
 
+    _settings: MistralLLMSettings
+
     def __init__(
         self,
         *,

src/pipecat/services/nvidia/llm.py

@@ -36,6 +36,8 @@ class NvidiaLLMService(OpenAILLMService):
     in token usage reporting between NIM (incremental) and OpenAI (final summary).
     """
 
+    _settings: NvidiaLLMSettings
+
     def __init__(
         self,
         *,

src/pipecat/services/ollama/llm.py

@@ -30,6 +30,8 @@ class OLLamaLLMService(OpenAILLMService):
     providing a compatible interface for running large language models locally.
     """
 
+    _settings: OllamaLLMSettings
+
     def __init__(
         self,
         *,

src/pipecat/services/openai/base_llm.py

@@ -53,11 +53,9 @@ class OpenAILLMSettings(LLMSettings):
 
     Parameters:
         max_completion_tokens: Maximum completion tokens to generate.
-        service_tier: Service tier to use (e.g., "auto", "flex", "priority").
     """
 
     max_completion_tokens: int | _NotGiven = field(default_factory=lambda: _NOT_GIVEN)
-    service_tier: str | _NotGiven = field(default_factory=lambda: _NOT_GIVEN)
 
 
 class BaseOpenAILLMService(LLMService):
@@ -118,6 +116,7 @@ class BaseOpenAILLMService(LLMService):
         organization=None,
         project=None,
         default_headers: Optional[Mapping[str, str]] = None,
+        service_tier: Optional[str] = None,
         params: Optional[InputParams] = None,
         settings: Optional[OpenAILLMSettings] = None,
         retry_timeout_secs: Optional[float] = 5.0,
@@ -138,6 +137,7 @@ class BaseOpenAILLMService(LLMService):
             organization: OpenAI organization ID.
             project: OpenAI project ID.
             default_headers: Additional HTTP headers to include in requests.
+            service_tier: Service tier to use (e.g., "auto", "flex", "priority").
             params: Input parameters for model configuration and behavior.
 
                 .. deprecated:: 0.0.105
@@ -161,7 +161,6 @@ class BaseOpenAILLMService(LLMService):
             top_k=None,
             max_tokens=NOT_GIVEN,
             max_completion_tokens=NOT_GIVEN,
-            service_tier=NOT_GIVEN,
             filter_incomplete_user_turns=False,
             user_turn_completion_config=None,
             extra={},
@@ -180,7 +179,6 @@ class BaseOpenAILLMService(LLMService):
             default_settings.top_p = params.top_p
             default_settings.max_tokens = params.max_tokens
             default_settings.max_completion_tokens = params.max_completion_tokens
-            default_settings.service_tier = params.service_tier
             if isinstance(params.extra, dict):
                 default_settings.extra = params.extra
 
@@ -192,6 +190,7 @@ class BaseOpenAILLMService(LLMService):
             settings=default_settings,
             **kwargs,
         )
+        self._service_tier = service_tier
         self._retry_timeout_secs = retry_timeout_secs
         self._retry_on_timeout = retry_on_timeout
         self._system_instruction = system_instruction
@@ -321,7 +320,7 @@ class BaseOpenAILLMService(LLMService):
             "top_p": self._settings.top_p,
             "max_tokens": self._settings.max_tokens,
             "max_completion_tokens": self._settings.max_completion_tokens,
-            "service_tier": self._settings.service_tier,
+            "service_tier": self._service_tier,
         }
 
         # Messages, tools, tool_choice

src/pipecat/services/openai/llm.py

@@ -76,6 +76,7 @@ class OpenAILLMService(BaseOpenAILLMService):
         self,
         *,
         model: Optional[str] = None,
+        service_tier: Optional[str] = None,
         params: Optional[BaseOpenAILLMService.InputParams] = None,
         settings: Optional[OpenAILLMSettings] = None,
         **kwargs,
@@ -88,6 +89,7 @@ class OpenAILLMService(BaseOpenAILLMService):
                 .. deprecated:: 0.0.105
                     Use ``settings=OpenAILLMSettings(model=...)`` instead.
 
+            service_tier: Service tier to use (e.g., "auto", "flex", "priority").
             params: Input parameters for model configuration.
 
                 .. deprecated:: 0.0.105
@@ -108,7 +110,6 @@ class OpenAILLMService(BaseOpenAILLMService):
             top_k=None,
             max_tokens=NOT_GIVEN,
             max_completion_tokens=NOT_GIVEN,
-            service_tier=NOT_GIVEN,
             filter_incomplete_user_turns=False,
             user_turn_completion_config=None,
             extra={},
@@ -119,6 +120,10 @@ class OpenAILLMService(BaseOpenAILLMService):
             _warn_deprecated_param("model", OpenAILLMSettings, "model")
             default_settings.model = model
 
+        # Handle service_tier from deprecated params
+        if params is not None and not settings and params.service_tier is not NOT_GIVEN:
+            service_tier = service_tier or params.service_tier
+
         # 3. Apply params overrides — only if settings not provided
         if params is not None:
             _warn_deprecated_param("params", OpenAILLMSettings)
@@ -130,7 +135,6 @@ class OpenAILLMService(BaseOpenAILLMService):
                 default_settings.top_p = params.top_p
                 default_settings.max_tokens = params.max_tokens
                 default_settings.max_completion_tokens = params.max_completion_tokens
-                default_settings.service_tier = params.service_tier
                 if isinstance(params.extra, dict):
                     default_settings.extra = params.extra
 
@@ -138,7 +142,7 @@ class OpenAILLMService(BaseOpenAILLMService):
         if settings is not None:
             default_settings.apply_update(settings)
 
-        super().__init__(settings=default_settings, **kwargs)
+        super().__init__(service_tier=service_tier, settings=default_settings, **kwargs)
 
     def create_context_aggregator(
         self,

src/pipecat/services/openai/realtime/llm.py

@@ -10,7 +10,7 @@ import base64
 import io
 import json
 import time
-from dataclasses import dataclass, field
+from dataclasses import dataclass
 from typing import Any, Optional
 
 from loguru import logger
@@ -59,7 +59,7 @@ from pipecat.processors.aggregators.openai_llm_context import (
 )
 from pipecat.processors.frame_processor import FrameDirection
 from pipecat.services.llm_service import FunctionCallFromLLM, LLMService
-from pipecat.services.settings import NOT_GIVEN, LLMSettings, _NotGiven, _warn_deprecated_param
+from pipecat.services.settings import LLMSettings, _warn_deprecated_param
 from pipecat.transcriptions.language import Language
 from pipecat.utils.time import time_now_iso8601
 from pipecat.utils.tracing.service_decorators import traced_openai_realtime, traced_stt
@@ -93,15 +93,9 @@ class CurrentAudioResponse:
 
 @dataclass
 class OpenAIRealtimeLLMSettings(LLMSettings):
-    """Settings for OpenAI Realtime LLM services.
+    """Settings for OpenAI Realtime LLM services."""
 
-    Parameters:
+    pass
-        session_properties: OpenAI Realtime session configuration.
-    """
-
-    session_properties: events.SessionProperties | _NotGiven = field(
-        default_factory=lambda: NOT_GIVEN
-    )
 
 
 class OpenAIRealtimeLLMService(LLMService):
@@ -145,15 +139,8 @@ class OpenAIRealtimeLLMService(LLMService):
             base_url: WebSocket base URL for the realtime API.
                 Defaults to "wss://api.openai.com/v1/realtime".
             session_properties: Configuration properties for the realtime session.
-
+                If None, uses default SessionProperties.
-                .. deprecated::
+            settings: Runtime-updatable settings for this service.
-                    Use ``settings=OpenAIRealtimeLLMSettings(session_properties=...)``
-                    instead.
-
-                These are session-level settings that can be updated during the session
-                (except for voice and model). If None, uses default SessionProperties.
-            settings: Realtime LLM settings. If provided together with deprecated
-                top-level parameters, the ``settings`` values take precedence.
             start_audio_paused: Whether to start with audio input paused. Defaults to False.
             start_video_paused: Whether to start with video input paused. Defaults to False.
             video_frame_detail: Detail level for video processing. Can be "auto", "low", or "high".
@@ -192,20 +179,14 @@ class OpenAIRealtimeLLMService(LLMService):
             seed=None,
             filter_incomplete_user_turns=False,
             user_turn_completion_config=None,
-            session_properties=events.SessionProperties(),
         )
 
         # 2. Apply direct init arg overrides (deprecated)
         if model is not None:
             _warn_deprecated_param("model", OpenAIRealtimeLLMSettings, "model")
             default_settings.model = model
-        if session_properties is not None:
-            _warn_deprecated_param(
-                "session_properties", OpenAIRealtimeLLMSettings, "session_properties"
-            )
-            default_settings.session_properties = session_properties
 
-        # 4. Apply settings delta (canonical API, always wins)
+        # 3. Apply settings delta (canonical API, always wins)
         if settings is not None:
             default_settings.apply_update(settings)
 
@@ -220,6 +201,7 @@ class OpenAIRealtimeLLMService(LLMService):
 
         self.api_key = api_key
         self.base_url = full_url
+        self._session_properties = session_properties or events.SessionProperties()
         self._audio_input_paused = start_audio_paused
         self._video_input_paused = start_video_paused
         self._video_frame_detail = video_frame_detail
@@ -282,12 +264,12 @@ class OpenAIRealtimeLLMService(LLMService):
 
     def _is_modality_enabled(self, modality: str) -> bool:
         """Check if a specific modality is enabled, "text" or "audio"."""
-        modalities = self._settings.session_properties.output_modalities or ["audio", "text"]
+        modalities = self._session_properties.output_modalities or ["audio", "text"]
         return modality in modalities
 
     def _get_enabled_modalities(self) -> list[str]:
         """Get the list of enabled modalities."""
-        modalities = self._settings.session_properties.output_modalities or ["audio", "text"]
+        modalities = self._session_properties.output_modalities or ["audio", "text"]
         # API only supports single modality responses: either ["text"] or ["audio"]
         if "audio" in modalities:
             return ["audio"]
@@ -360,9 +342,9 @@ class OpenAIRealtimeLLMService(LLMService):
         # None and False are different. Check for False. None means we're using OpenAI's
         # built-in turn detection defaults.
         turn_detection_disabled = (
-            self._settings.session_properties.audio
+            self._session_properties.audio
-            and self._settings.session_properties.audio.input
+            and self._session_properties.audio.input
-            and self._settings.session_properties.audio.input.turn_detection is False
+            and self._session_properties.audio.input.turn_detection is False
         )
         if turn_detection_disabled:
             await self.send_client_event(events.InputAudioBufferClearEvent())
@@ -382,9 +364,9 @@ class OpenAIRealtimeLLMService(LLMService):
         # None and False are different. Check for False. None means we're using OpenAI's
         # built-in turn detection defaults.
         turn_detection_disabled = (
-            self._settings.session_properties.audio
+            self._session_properties.audio
-            and self._settings.session_properties.audio.input
+            and self._session_properties.audio.input
-            and self._settings.session_properties.audio.input.turn_detection is False
+            and self._session_properties.audio.input.turn_detection is False
         )
         if turn_detection_disabled:
             await self.send_client_event(events.InputAudioBufferCommitEvent())
@@ -457,7 +439,7 @@ class OpenAIRealtimeLLMService(LLMService):
         # directly. The frame.delta path falls through to super, which calls
         # _update_settings → our override handles the rest.
         if isinstance(frame, LLMUpdateSettingsFrame) and frame.delta is None:
-            self._settings.session_properties = events.SessionProperties(**frame.settings)
+            self._session_properties = events.SessionProperties(**frame.settings)
             await self._send_session_update()
             await self.push_frame(frame, direction)
             return
@@ -576,15 +558,13 @@ class OpenAIRealtimeLLMService(LLMService):
             await self.push_error(error_msg=f"Error sending client event: {e}", exception=e)
 
     async def _update_settings(self, delta):
-        """Apply a settings delta, sending a session update if needed."""
+        """Apply a settings delta."""
         changed = await super()._update_settings(delta)
-        if "session_properties" in changed:
+        self._warn_unhandled_updated_settings(changed.keys())
-            await self._send_session_update()
-        self._warn_unhandled_updated_settings(changed.keys() - {"session_properties"})
         return changed
 
     async def _send_session_update(self):
-        settings = self._settings.session_properties
+        settings = self._session_properties
         adapter: OpenAIRealtimeLLMAdapter = self.get_llm_adapter()
 
         if self._context:

src/pipecat/services/openai_realtime_beta/azure.py

@@ -42,6 +42,8 @@ class AzureRealtimeBetaLLMService(OpenAIRealtimeBetaLLMService):
     real-time audio and text communication capabilities as the base OpenAI service.
     """
 
+    _settings: AzureRealtimeBetaLLMSettings
+
     def __init__(
         self,
         *,

src/pipecat/services/openai_realtime_beta/openai.py

@@ -10,7 +10,7 @@ import base64
 import json
 import time
 import warnings
-from dataclasses import dataclass, field
+from dataclasses import dataclass
 from typing import Optional
 
 from loguru import logger
@@ -54,7 +54,7 @@ from pipecat.processors.aggregators.openai_llm_context import (
 from pipecat.processors.frame_processor import FrameDirection
 from pipecat.services.llm_service import FunctionCallFromLLM, LLMService
 from pipecat.services.openai.llm import OpenAIContextAggregatorPair
-from pipecat.services.settings import NOT_GIVEN, LLMSettings, _NotGiven, _warn_deprecated_param
+from pipecat.services.settings import LLMSettings, _warn_deprecated_param
 from pipecat.transcriptions.language import Language
 from pipecat.utils.time import time_now_iso8601
 from pipecat.utils.tracing.service_decorators import traced_openai_realtime, traced_stt
@@ -94,15 +94,9 @@ class CurrentAudioResponse:
 
 @dataclass
 class OpenAIRealtimeBetaLLMSettings(LLMSettings):
-    """Settings for OpenAI Realtime Beta LLM services.
+    """Settings for OpenAI Realtime Beta LLM services."""
 
-    Parameters:
+    pass
-        session_properties: OpenAI Realtime session configuration.
-    """
-
-    session_properties: events.SessionProperties | _NotGiven = field(
-        default_factory=lambda: NOT_GIVEN
-    )
 
 
 class OpenAIRealtimeBetaLLMService(LLMService):
@@ -146,14 +140,8 @@ class OpenAIRealtimeBetaLLMService(LLMService):
             base_url: WebSocket base URL for the realtime API.
                 Defaults to "wss://api.openai.com/v1/realtime".
             session_properties: Configuration properties for the realtime session.
-
-                .. deprecated::
-                    Use ``settings=OpenAIRealtimeBetaLLMSettings(session_properties=...)``
-                    instead.
-
                 If None, uses default SessionProperties.
-            settings: Realtime Beta LLM settings. If provided together with deprecated
+            settings: Runtime-updatable settings for this service.
-                top-level parameters, the ``settings`` values take precedence.
             start_audio_paused: Whether to start with audio input paused. Defaults to False.
             send_transcription_frames: Whether to emit transcription frames. Defaults to True.
             **kwargs: Additional arguments passed to parent LLMService.
@@ -179,20 +167,13 @@ class OpenAIRealtimeBetaLLMService(LLMService):
             seed=None,
             filter_incomplete_user_turns=False,
             user_turn_completion_config=None,
-            session_properties=events.SessionProperties(),
         )
 
         # 2. Apply direct init arg overrides (deprecated)
         if model is not None:
             _warn_deprecated_param("model", OpenAIRealtimeBetaLLMSettings, "model")
             default_settings.model = model
-        if session_properties is not None:
+        # 3. Apply settings delta (canonical API, always wins)
-            _warn_deprecated_param(
-                "session_properties", OpenAIRealtimeBetaLLMSettings, "session_properties"
-            )
-            default_settings.session_properties = session_properties
-
-        # 4. Apply settings delta (canonical API, always wins)
         if settings is not None:
             default_settings.apply_update(settings)
 
@@ -205,6 +186,7 @@ class OpenAIRealtimeBetaLLMService(LLMService):
 
         self.api_key = api_key
         self.base_url = full_url
+        self._session_properties = session_properties or events.SessionProperties()
         self._audio_input_paused = start_audio_paused
         self._send_transcription_frames = send_transcription_frames
         self._websocket = None
@@ -243,12 +225,12 @@ class OpenAIRealtimeBetaLLMService(LLMService):
 
     def _is_modality_enabled(self, modality: str) -> bool:
         """Check if a specific modality is enabled, "text" or "audio"."""
-        modalities = self._settings.session_properties.modalities or ["audio", "text"]
+        modalities = self._session_properties.modalities or ["audio", "text"]
         return modality in modalities
 
     def _get_enabled_modalities(self) -> list[str]:
         """Get the list of enabled modalities."""
-        return self._settings.session_properties.modalities or ["audio", "text"]
+        return self._session_properties.modalities or ["audio", "text"]
 
     async def retrieve_conversation_item(self, item_id: str):
         """Retrieve a conversation item by ID from the server.
@@ -315,7 +297,7 @@ class OpenAIRealtimeBetaLLMService(LLMService):
     async def _handle_interruption(self):
         # None and False are different. Check for False. None means we're using OpenAI's
         # built-in turn detection defaults.
-        if self._settings.session_properties.turn_detection is False:
+        if self._session_properties.turn_detection is False:
             await self.send_client_event(events.InputAudioBufferClearEvent())
             await self.send_client_event(events.ResponseCancelEvent())
         await self._truncate_current_audio_response()
@@ -332,7 +314,7 @@ class OpenAIRealtimeBetaLLMService(LLMService):
     async def _handle_user_stopped_speaking(self, frame):
         # None and False are different. Check for False. None means we're using OpenAI's
         # built-in turn detection defaults.
-        if self._settings.session_properties.turn_detection is False:
+        if self._session_properties.turn_detection is False:
             await self.send_client_event(events.InputAudioBufferCommitEvent())
             await self.send_client_event(events.ResponseCreateEvent())
 
@@ -403,7 +385,7 @@ class OpenAIRealtimeBetaLLMService(LLMService):
         # directly. The frame.delta path falls through to super, which calls
         # _update_settings → our override handles the rest.
         if isinstance(frame, LLMUpdateSettingsFrame) and frame.delta is None:
-            self._settings.session_properties = events.SessionProperties(**frame.settings)
+            self._session_properties = events.SessionProperties(**frame.settings)
             await self._send_session_update()
             await self.push_frame(frame, direction)
             return
@@ -520,14 +502,13 @@ class OpenAIRealtimeBetaLLMService(LLMService):
             await self.push_error(error_msg=f"Error sending client event: {e}", exception=e)
 
     async def _update_settings(self, delta):
-        """Apply a settings delta, sending a session update if needed."""
+        """Apply a settings delta."""
         changed = await super()._update_settings(delta)
-        if "session_properties" in changed:
+        self._warn_unhandled_updated_settings(changed.keys())
-            await self._send_session_update()
         return changed
 
     async def _send_session_update(self):
-        settings = self._settings.session_properties
+        settings = self._session_properties
         # tools given in the context override the tools in the session properties
         if self._context and self._context.tools:
             settings.tools = self._context.tools

src/pipecat/services/openpipe/llm.py

@@ -43,6 +43,8 @@ class OpenPipeLLMService(OpenAILLMService):
     for model training and evaluation.
     """
 
+    _settings: OpenPipeLLMSettings
+
     def __init__(
         self,
         *,

src/pipecat/services/openrouter/llm.py

@@ -34,6 +34,8 @@ class OpenRouterLLMService(OpenAILLMService):
     maintaining full compatibility with OpenAI's interface and functionality.
     """
 
+    _settings: OpenRouterLLMSettings
+
     def __init__(
         self,
         *,

src/pipecat/services/perplexity/llm.py

@@ -38,6 +38,8 @@ class PerplexityLLMService(OpenAILLMService):
     in token usage reporting between Perplexity (incremental) and OpenAI (final summary).
     """
 
+    _settings: PerplexityLLMSettings
+
     def __init__(
         self,
         *,

src/pipecat/services/qwen/llm.py

@@ -30,6 +30,8 @@ class QwenLLMService(OpenAILLMService):
     maintaining full compatibility with OpenAI's interface and functionality.
     """
 
+    _settings: QwenLLMSettings
+
     def __init__(
         self,
         *,

src/pipecat/services/sambanova/llm.py

@@ -42,6 +42,8 @@ class SambaNovaLLMService(OpenAILLMService):  # type: ignore
     maintaining full compatibility with OpenAI's interface and functionality.
     """
 
+    _settings: SambaNovaLLMSettings
+
     def __init__(
         self,
         *,

src/pipecat/services/together/llm.py

@@ -30,6 +30,8 @@ class TogetherLLMService(OpenAILLMService):
     maintaining full compatibility with OpenAI's interface and functionality.
     """
 
+    _settings: TogetherLLMSettings
+
     def __init__(
         self,
         *,