diff --git a/litellm/llms/gemini/audio_transcription/realtime_transformation.py b/litellm/llms/gemini/audio_transcription/realtime_transformation.py new file mode 100644 index 00000000000..22d13c41b56 --- /dev/null +++ b/litellm/llms/gemini/audio_transcription/realtime_transformation.py @@ -0,0 +1,8 @@ +from litellm.types.llms.openai import OpenAIAudioTranscriptionOptionalParams + +from .transformation import GeminiAudioTranscriptionConfig + + +class GeminiRealtimeAudioTranscriptionConfig(GeminiAudioTranscriptionConfig): + def get_supported_openai_params(self, model: str) -> list[OpenAIAudioTranscriptionOptionalParams]: + return ["language", "keywords"] diff --git a/litellm/model_prices_and_context_window_backup.json b/litellm/model_prices_and_context_window_backup.json index 67b851a4093..3e85e41b7b2 100644 --- a/litellm/model_prices_and_context_window_backup.json +++ b/litellm/model_prices_and_context_window_backup.json @@ -62738,6 +62738,7 @@ "supports_function_calling": false }, "gemini/gemini-3.5-transcribe-live": { + "audio_transcription_config": "gemini_realtime_transcribe", "input_cost_per_audio_token": 3.5e-06, "input_cost_per_token": 3.5e-06, "litellm_provider": "gemini", diff --git a/litellm/utils.py b/litellm/utils.py index 6f77ce1a5bf..a667b60f7c5 100644 --- a/litellm/utils.py +++ b/litellm/utils.py @@ -8951,6 +8951,15 @@ class ProviderConfigManager: ) return VertexAIAudioTranscriptionConfig() + elif ( + litellm.LlmProviders.GEMINI == provider + and model_cost_entry.get("audio_transcription_config") == "gemini_realtime_transcribe" + ): + from litellm.llms.gemini.audio_transcription.realtime_transformation import ( + GeminiRealtimeAudioTranscriptionConfig, + ) + + return GeminiRealtimeAudioTranscriptionConfig() elif ( litellm.LlmProviders.GEMINI == provider and model_cost_entry.get("audio_transcription_config") == "gemini_transcribe" diff --git a/model_prices_and_context_window.json b/model_prices_and_context_window.json index 67b851a4093..3e85e41b7b2 100644 --- a/model_prices_and_context_window.json +++ b/model_prices_and_context_window.json @@ -62738,6 +62738,7 @@ "supports_function_calling": false }, "gemini/gemini-3.5-transcribe-live": { + "audio_transcription_config": "gemini_realtime_transcribe", "input_cost_per_audio_token": 3.5e-06, "input_cost_per_token": 3.5e-06, "litellm_provider": "gemini", diff --git a/tests/litellm_utils_tests/test_utils.py b/tests/litellm_utils_tests/test_utils.py index f3016f34df1..4760effc9d9 100644 --- a/tests/litellm_utils_tests/test_utils.py +++ b/tests/litellm_utils_tests/test_utils.py @@ -2148,18 +2148,28 @@ def test_gemini_model_info_reports_transcription_params(monkeypatch): assert "keywords" in model_info["supported_openai_params"] -def test_gemini_live_transcription_dispatch_has_no_batch_params(): +def test_gemini_live_transcription_dispatch_returns_realtime_params(monkeypatch): from litellm.litellm_core_utils.get_supported_openai_params import ( get_supported_openai_params, ) + monkeypatch.setitem( + litellm.model_cost, + "gemini/gemini-3.5-transcribe-live", + { + "audio_transcription_config": "gemini_realtime_transcribe", + "mode": "audio_transcription", + "supported_endpoints": ["/v1/realtime"], + }, + ) + params = get_supported_openai_params( model="gemini-3.5-transcribe-live", custom_llm_provider="gemini", request_type="transcription", ) - assert params is None + assert params == ["language", "keywords"] @pytest.mark.parametrize( diff --git a/tests/test_litellm/litellm_core_utils/test_get_supported_openai_params.py b/tests/test_litellm/litellm_core_utils/test_get_supported_openai_params.py index d40cf46420c..dbe61c6cb34 100644 --- a/tests/test_litellm/litellm_core_utils/test_get_supported_openai_params.py +++ b/tests/test_litellm/litellm_core_utils/test_get_supported_openai_params.py @@ -27,14 +27,24 @@ def test_gemini_transcription_dispatch_returns_batch_params(monkeypatch): assert "timestamp_granularities" in params -def test_gemini_live_transcription_dispatch_does_not_use_batch_params(): +def test_gemini_live_transcription_dispatch_returns_realtime_params(monkeypatch): + monkeypatch.setitem( + litellm.model_cost, + "gemini/gemini-3.5-transcribe-live", + { + "audio_transcription_config": "gemini_realtime_transcribe", + "mode": "audio_transcription", + "supported_endpoints": ["/v1/realtime"], + }, + ) + params = get_supported_openai_params( model="gemini-3.5-transcribe-live", custom_llm_provider="gemini", request_type="transcription", ) - assert params is None + assert params == ["language", "keywords"] def test_base_model_label_does_not_strip_bedrock_tools(): diff --git a/tests/unit/llms/gemini/audio_transcription/test_gemini_audio_transcription_transformation.py b/tests/unit/llms/gemini/audio_transcription/test_gemini_audio_transcription_transformation.py index de3a1653e6b..87fbfd2b2ae 100644 --- a/tests/unit/llms/gemini/audio_transcription/test_gemini_audio_transcription_transformation.py +++ b/tests/unit/llms/gemini/audio_transcription/test_gemini_audio_transcription_transformation.py @@ -5,9 +5,10 @@ import httpx import pytest import litellm -from litellm.llms.gemini.audio_transcription.transformation import ( - GeminiAudioTranscriptionConfig, +from litellm.llms.gemini.audio_transcription.realtime_transformation import ( + GeminiRealtimeAudioTranscriptionConfig, ) +from litellm.llms.gemini.audio_transcription.transformation import GeminiAudioTranscriptionConfig from litellm.llms.gemini.common_utils import GeminiError from litellm.types.utils import LlmProviders from litellm.utils import ProviderConfigManager, get_optional_params_transcription @@ -72,11 +73,12 @@ def test_provider_config_manager_returns_gemini_config(): assert isinstance(provider_config, GeminiAudioTranscriptionConfig) -def test_provider_config_manager_does_not_use_batch_config_for_gemini_live(): +def test_provider_config_manager_uses_realtime_config_for_gemini_live(): provider_config = ProviderConfigManager.get_provider_audio_transcription_config( model="gemini-3.5-transcribe-live", provider=LlmProviders.GEMINI ) - assert provider_config is None + assert isinstance(provider_config, GeminiRealtimeAudioTranscriptionConfig) + assert provider_config.get_supported_openai_params("gemini-3.5-transcribe-live") == ["language", "keywords"] class TestSupportedParams: