diff --git a/litellm/llms/azure/audio_transcriptions.py b/litellm/llms/azure/audio_transcriptions.py index 754c319d0d0..e5675d4c82f 100644 --- a/litellm/llms/azure/audio_transcriptions.py +++ b/litellm/llms/azure/audio_transcriptions.py @@ -43,8 +43,15 @@ class AzureAudioTranscription(AzureChatCompletion): ) -> TranscriptionResponse | Coroutine[Any, Any, TranscriptionResponse]: data: Final = {"model": model, "file": audio_file, **optional_params} sdk_data: Final = sdk_compatible_transcription_request_data(data) + model_info: Final = litellm.model_cost.get(f"azure/{model}") + provider_specific_entry: Final = model_info.get("provider_specific_entry") if model_info is not None else None + requires_deployment_api: Final = model_info is None or ( + provider_specific_entry is not None and provider_specific_entry.get("transcription_deployment_api") == 1 + ) resolved_api_version: Final = ( - litellm.AZURE_DEFAULT_API_VERSION if api_version in ("v1", "latest", "preview") else api_version + litellm.AZURE_DEFAULT_API_VERSION + if requires_deployment_api and api_version in ("v1", "latest", "preview") + else api_version ) if atranscription is True: diff --git a/litellm/model_prices_and_context_window_backup.json b/litellm/model_prices_and_context_window_backup.json index 0b2117a627e..82c2d9519b1 100644 --- a/litellm/model_prices_and_context_window_backup.json +++ b/litellm/model_prices_and_context_window_backup.json @@ -234,6 +234,7 @@ "mode": "audio_transcription", "provider_specific_entry": { "realtime_ga_only": 1, + "transcription_deployment_api": 1, "transcription_json_only": 1 }, "source": "https://techcommunity.microsoft.com/blog/azure-ai-foundry-blog/introducing-gpt-transcribe-and-gpt-live-transcribe-in-microsoft-foundry/4541740", @@ -25728,6 +25729,7 @@ }, "gemini-3-pro-image": { "cache_read_input_token_cost": 2e-07, + "cache_read_input_token_cost_batches": 1e-07, "cache_read_input_token_cost_above_200k_tokens": 4e-07, "cache_read_input_token_cost_above_200k_tokens_priority": 7.2e-07, "cache_read_input_token_cost_flex": 1e-07, @@ -25822,6 +25824,7 @@ }, "gemini-3.1-flash-image": { "cache_read_input_token_cost": 5e-08, + "cache_read_input_token_cost_batches": 2.5e-08, "cache_read_input_token_cost_flex": 2.5e-08, "deprecation_date": "2027-05-28", "input_cost_per_image": 0.00056, @@ -25905,6 +25908,7 @@ }, "gemini-3.1-flash-lite-image": { "cache_read_input_token_cost": 2.5e-08, + "cache_read_input_token_cost_batches": 1.25e-08, "cache_read_input_token_cost_flex": 1.25e-08, "input_cost_per_image": 0.00028, "input_cost_per_token": 2.5e-07, @@ -25997,6 +26001,7 @@ "cache_read_input_audio_token_cost": 5e-08, "deprecation_date": "2027-05-07", "cache_read_input_token_cost": 2.5e-08, + "cache_read_input_token_cost_batches": 1.25e-08, "cache_read_input_token_cost_flex": 1.25e-08, "cache_read_input_token_cost_priority": 4.5e-08, "input_cost_per_audio_token": 5e-07, @@ -26056,6 +26061,7 @@ "gemini-3.5-flash-lite": { "deprecation_date": "2027-07-21", "cache_read_input_token_cost": 3e-08, + "cache_read_input_token_cost_batches": 1.5e-08, "cache_read_input_token_cost_flex": 1.5e-08, "cache_read_input_token_cost_priority": 5.4e-08, "input_cost_per_token": 3e-07, @@ -26112,6 +26118,7 @@ }, "deep-research-pro-preview-12-2025": { "cache_read_input_token_cost": 2e-07, + "cache_read_input_token_cost_batches": 1e-07, "input_cost_per_image": 0.0011, "input_cost_per_token": 2e-06, "input_cost_per_token_batches": 1e-06, @@ -26717,6 +26724,7 @@ "prompt_cache_min_tokens": 4096, "deprecation_date": "2027-05-19", "cache_read_input_token_cost": 1.5e-07, + "cache_read_input_token_cost_batches": 7.5e-08, "input_cost_per_token": 1.5e-06, "input_cost_per_audio_token": 1.5e-06, "litellm_provider": "vertex_ai", @@ -26776,6 +26784,7 @@ "vertex_ai/gemini-3.6-flash": { "prompt_cache_min_tokens": 4096, "cache_read_input_token_cost": 7.5e-08, + "cache_read_input_token_cost_batches": 3.75e-08, "cache_read_input_token_cost_flex": 3.75e-08, "input_cost_per_token": 7.5e-07, "input_cost_per_token_batches": 3.75e-07, @@ -26833,6 +26842,7 @@ "vertex_ai/gemini-3.7-flash": { "prompt_cache_min_tokens": 4096, "cache_read_input_token_cost": 7.5e-08, + "cache_read_input_token_cost_batches": 3.75e-08, "cache_read_input_token_cost_flex": 3.75e-08, "input_cost_per_token": 7.5e-07, "input_cost_per_token_batches": 3.75e-07, @@ -26891,6 +26901,7 @@ "vertex_ai/gemini-3.8-flash": { "prompt_cache_min_tokens": 4096, "cache_read_input_token_cost": 7.5e-08, + "cache_read_input_token_cost_batches": 3.75e-08, "cache_read_input_token_cost_flex": 3.75e-08, "input_cost_per_token": 7.5e-07, "input_cost_per_token_batches": 3.75e-07, @@ -28638,6 +28649,7 @@ "prompt_cache_min_tokens": 4096, "deprecation_date": "2027-05-19", "cache_read_input_token_cost": 1.5e-07, + "cache_read_input_token_cost_batches": 7.5e-08, "input_cost_per_audio_token": 1.5e-06, "input_cost_per_token": 1.5e-06, "litellm_provider": "vertex_ai-language-models", @@ -28697,6 +28709,7 @@ "gemini-3.6-flash": { "prompt_cache_min_tokens": 4096, "cache_read_input_token_cost": 7.5e-08, + "cache_read_input_token_cost_batches": 3.75e-08, "cache_read_input_token_cost_flex": 3.75e-08, "input_cost_per_token": 7.5e-07, "input_cost_per_token_batches": 3.75e-07, @@ -28754,6 +28767,7 @@ "gemini-3.7-flash": { "prompt_cache_min_tokens": 4096, "cache_read_input_token_cost": 7.5e-08, + "cache_read_input_token_cost_batches": 3.75e-08, "cache_read_input_token_cost_flex": 3.75e-08, "input_cost_per_token": 7.5e-07, "input_cost_per_token_batches": 3.75e-07, @@ -28812,6 +28826,7 @@ "gemini-3.8-flash": { "prompt_cache_min_tokens": 4096, "cache_read_input_token_cost": 7.5e-08, + "cache_read_input_token_cost_batches": 3.75e-08, "cache_read_input_token_cost_flex": 3.75e-08, "input_cost_per_token": 7.5e-07, "input_cost_per_token_batches": 3.75e-07, @@ -33499,6 +33514,7 @@ }, "gpt-5-2025-08-07": { "cache_read_input_token_cost": 1.25e-07, + "cache_read_input_token_cost_batches": 6.25e-08, "cache_read_input_token_cost_flex": 6.25e-08, "cache_read_input_token_cost_priority": 2.5e-07, "deprecation_date": "2026-12-11", @@ -33679,6 +33695,7 @@ }, "gpt-5-mini-2025-08-07": { "cache_read_input_token_cost": 2.5e-08, + "cache_read_input_token_cost_batches": 1.25e-08, "cache_read_input_token_cost_flex": 1.25e-08, "cache_read_input_token_cost_priority": 4.5e-08, "deprecation_date": "2026-12-11", @@ -33779,6 +33796,7 @@ }, "gpt-5-nano-2025-08-07": { "cache_read_input_token_cost": 5e-09, + "cache_read_input_token_cost_batches": 2.5e-09, "cache_read_input_token_cost_flex": 2.5e-09, "deprecation_date": "2026-12-11", "input_cost_per_token": 5e-08, @@ -48117,6 +48135,7 @@ }, "vertex_ai/gemini-3-pro-image": { "cache_read_input_token_cost": 2e-07, + "cache_read_input_token_cost_batches": 1e-07, "cache_read_input_token_cost_above_200k_tokens": 4e-07, "cache_read_input_token_cost_above_200k_tokens_priority": 7.2e-07, "cache_read_input_token_cost_flex": 1e-07, @@ -48163,6 +48182,7 @@ }, "vertex_ai/gemini-3.1-flash-image": { "cache_read_input_token_cost": 5e-08, + "cache_read_input_token_cost_batches": 2.5e-08, "cache_read_input_token_cost_flex": 2.5e-08, "deprecation_date": "2027-05-28", "input_cost_per_image": 0.00056, @@ -48198,6 +48218,7 @@ }, "vertex_ai/gemini-3.1-flash-lite-image": { "cache_read_input_token_cost": 2.5e-08, + "cache_read_input_token_cost_batches": 1.25e-08, "cache_read_input_token_cost_flex": 1.25e-08, "input_cost_per_image": 0.00028, "input_cost_per_token": 2.5e-07, @@ -48290,6 +48311,7 @@ "cache_read_input_audio_token_cost": 5e-08, "deprecation_date": "2027-05-07", "cache_read_input_token_cost": 2.5e-08, + "cache_read_input_token_cost_batches": 1.25e-08, "cache_read_input_token_cost_flex": 1.25e-08, "cache_read_input_token_cost_priority": 4.5e-08, "input_cost_per_audio_token": 5e-07, @@ -48350,6 +48372,7 @@ "vertex_ai/gemini-3.5-flash-lite": { "deprecation_date": "2027-07-21", "cache_read_input_token_cost": 3e-08, + "cache_read_input_token_cost_batches": 1.5e-08, "cache_read_input_token_cost_flex": 1.5e-08, "cache_read_input_token_cost_priority": 5.4e-08, "input_cost_per_token": 3e-07, @@ -48407,6 +48430,7 @@ }, "vertex_ai/deep-research-pro-preview-12-2025": { "cache_read_input_token_cost": 2e-07, + "cache_read_input_token_cost_batches": 1e-07, "input_cost_per_image": 0.0011, "input_cost_per_token": 2e-06, "input_cost_per_token_batches": 1e-06, diff --git a/model_prices_and_context_window.json b/model_prices_and_context_window.json index 0b2117a627e..82c2d9519b1 100644 --- a/model_prices_and_context_window.json +++ b/model_prices_and_context_window.json @@ -234,6 +234,7 @@ "mode": "audio_transcription", "provider_specific_entry": { "realtime_ga_only": 1, + "transcription_deployment_api": 1, "transcription_json_only": 1 }, "source": "https://techcommunity.microsoft.com/blog/azure-ai-foundry-blog/introducing-gpt-transcribe-and-gpt-live-transcribe-in-microsoft-foundry/4541740", @@ -25728,6 +25729,7 @@ }, "gemini-3-pro-image": { "cache_read_input_token_cost": 2e-07, + "cache_read_input_token_cost_batches": 1e-07, "cache_read_input_token_cost_above_200k_tokens": 4e-07, "cache_read_input_token_cost_above_200k_tokens_priority": 7.2e-07, "cache_read_input_token_cost_flex": 1e-07, @@ -25822,6 +25824,7 @@ }, "gemini-3.1-flash-image": { "cache_read_input_token_cost": 5e-08, + "cache_read_input_token_cost_batches": 2.5e-08, "cache_read_input_token_cost_flex": 2.5e-08, "deprecation_date": "2027-05-28", "input_cost_per_image": 0.00056, @@ -25905,6 +25908,7 @@ }, "gemini-3.1-flash-lite-image": { "cache_read_input_token_cost": 2.5e-08, + "cache_read_input_token_cost_batches": 1.25e-08, "cache_read_input_token_cost_flex": 1.25e-08, "input_cost_per_image": 0.00028, "input_cost_per_token": 2.5e-07, @@ -25997,6 +26001,7 @@ "cache_read_input_audio_token_cost": 5e-08, "deprecation_date": "2027-05-07", "cache_read_input_token_cost": 2.5e-08, + "cache_read_input_token_cost_batches": 1.25e-08, "cache_read_input_token_cost_flex": 1.25e-08, "cache_read_input_token_cost_priority": 4.5e-08, "input_cost_per_audio_token": 5e-07, @@ -26056,6 +26061,7 @@ "gemini-3.5-flash-lite": { "deprecation_date": "2027-07-21", "cache_read_input_token_cost": 3e-08, + "cache_read_input_token_cost_batches": 1.5e-08, "cache_read_input_token_cost_flex": 1.5e-08, "cache_read_input_token_cost_priority": 5.4e-08, "input_cost_per_token": 3e-07, @@ -26112,6 +26118,7 @@ }, "deep-research-pro-preview-12-2025": { "cache_read_input_token_cost": 2e-07, + "cache_read_input_token_cost_batches": 1e-07, "input_cost_per_image": 0.0011, "input_cost_per_token": 2e-06, "input_cost_per_token_batches": 1e-06, @@ -26717,6 +26724,7 @@ "prompt_cache_min_tokens": 4096, "deprecation_date": "2027-05-19", "cache_read_input_token_cost": 1.5e-07, + "cache_read_input_token_cost_batches": 7.5e-08, "input_cost_per_token": 1.5e-06, "input_cost_per_audio_token": 1.5e-06, "litellm_provider": "vertex_ai", @@ -26776,6 +26784,7 @@ "vertex_ai/gemini-3.6-flash": { "prompt_cache_min_tokens": 4096, "cache_read_input_token_cost": 7.5e-08, + "cache_read_input_token_cost_batches": 3.75e-08, "cache_read_input_token_cost_flex": 3.75e-08, "input_cost_per_token": 7.5e-07, "input_cost_per_token_batches": 3.75e-07, @@ -26833,6 +26842,7 @@ "vertex_ai/gemini-3.7-flash": { "prompt_cache_min_tokens": 4096, "cache_read_input_token_cost": 7.5e-08, + "cache_read_input_token_cost_batches": 3.75e-08, "cache_read_input_token_cost_flex": 3.75e-08, "input_cost_per_token": 7.5e-07, "input_cost_per_token_batches": 3.75e-07, @@ -26891,6 +26901,7 @@ "vertex_ai/gemini-3.8-flash": { "prompt_cache_min_tokens": 4096, "cache_read_input_token_cost": 7.5e-08, + "cache_read_input_token_cost_batches": 3.75e-08, "cache_read_input_token_cost_flex": 3.75e-08, "input_cost_per_token": 7.5e-07, "input_cost_per_token_batches": 3.75e-07, @@ -28638,6 +28649,7 @@ "prompt_cache_min_tokens": 4096, "deprecation_date": "2027-05-19", "cache_read_input_token_cost": 1.5e-07, + "cache_read_input_token_cost_batches": 7.5e-08, "input_cost_per_audio_token": 1.5e-06, "input_cost_per_token": 1.5e-06, "litellm_provider": "vertex_ai-language-models", @@ -28697,6 +28709,7 @@ "gemini-3.6-flash": { "prompt_cache_min_tokens": 4096, "cache_read_input_token_cost": 7.5e-08, + "cache_read_input_token_cost_batches": 3.75e-08, "cache_read_input_token_cost_flex": 3.75e-08, "input_cost_per_token": 7.5e-07, "input_cost_per_token_batches": 3.75e-07, @@ -28754,6 +28767,7 @@ "gemini-3.7-flash": { "prompt_cache_min_tokens": 4096, "cache_read_input_token_cost": 7.5e-08, + "cache_read_input_token_cost_batches": 3.75e-08, "cache_read_input_token_cost_flex": 3.75e-08, "input_cost_per_token": 7.5e-07, "input_cost_per_token_batches": 3.75e-07, @@ -28812,6 +28826,7 @@ "gemini-3.8-flash": { "prompt_cache_min_tokens": 4096, "cache_read_input_token_cost": 7.5e-08, + "cache_read_input_token_cost_batches": 3.75e-08, "cache_read_input_token_cost_flex": 3.75e-08, "input_cost_per_token": 7.5e-07, "input_cost_per_token_batches": 3.75e-07, @@ -33499,6 +33514,7 @@ }, "gpt-5-2025-08-07": { "cache_read_input_token_cost": 1.25e-07, + "cache_read_input_token_cost_batches": 6.25e-08, "cache_read_input_token_cost_flex": 6.25e-08, "cache_read_input_token_cost_priority": 2.5e-07, "deprecation_date": "2026-12-11", @@ -33679,6 +33695,7 @@ }, "gpt-5-mini-2025-08-07": { "cache_read_input_token_cost": 2.5e-08, + "cache_read_input_token_cost_batches": 1.25e-08, "cache_read_input_token_cost_flex": 1.25e-08, "cache_read_input_token_cost_priority": 4.5e-08, "deprecation_date": "2026-12-11", @@ -33779,6 +33796,7 @@ }, "gpt-5-nano-2025-08-07": { "cache_read_input_token_cost": 5e-09, + "cache_read_input_token_cost_batches": 2.5e-09, "cache_read_input_token_cost_flex": 2.5e-09, "deprecation_date": "2026-12-11", "input_cost_per_token": 5e-08, @@ -48117,6 +48135,7 @@ }, "vertex_ai/gemini-3-pro-image": { "cache_read_input_token_cost": 2e-07, + "cache_read_input_token_cost_batches": 1e-07, "cache_read_input_token_cost_above_200k_tokens": 4e-07, "cache_read_input_token_cost_above_200k_tokens_priority": 7.2e-07, "cache_read_input_token_cost_flex": 1e-07, @@ -48163,6 +48182,7 @@ }, "vertex_ai/gemini-3.1-flash-image": { "cache_read_input_token_cost": 5e-08, + "cache_read_input_token_cost_batches": 2.5e-08, "cache_read_input_token_cost_flex": 2.5e-08, "deprecation_date": "2027-05-28", "input_cost_per_image": 0.00056, @@ -48198,6 +48218,7 @@ }, "vertex_ai/gemini-3.1-flash-lite-image": { "cache_read_input_token_cost": 2.5e-08, + "cache_read_input_token_cost_batches": 1.25e-08, "cache_read_input_token_cost_flex": 1.25e-08, "input_cost_per_image": 0.00028, "input_cost_per_token": 2.5e-07, @@ -48290,6 +48311,7 @@ "cache_read_input_audio_token_cost": 5e-08, "deprecation_date": "2027-05-07", "cache_read_input_token_cost": 2.5e-08, + "cache_read_input_token_cost_batches": 1.25e-08, "cache_read_input_token_cost_flex": 1.25e-08, "cache_read_input_token_cost_priority": 4.5e-08, "input_cost_per_audio_token": 5e-07, @@ -48350,6 +48372,7 @@ "vertex_ai/gemini-3.5-flash-lite": { "deprecation_date": "2027-07-21", "cache_read_input_token_cost": 3e-08, + "cache_read_input_token_cost_batches": 1.5e-08, "cache_read_input_token_cost_flex": 1.5e-08, "cache_read_input_token_cost_priority": 5.4e-08, "input_cost_per_token": 3e-07, @@ -48407,6 +48430,7 @@ }, "vertex_ai/deep-research-pro-preview-12-2025": { "cache_read_input_token_cost": 2e-07, + "cache_read_input_token_cost_batches": 1e-07, "input_cost_per_image": 0.0011, "input_cost_per_token": 2e-06, "input_cost_per_token_batches": 1e-06, diff --git a/tests/test_litellm/llms/azure/test_audio_transcriptions.py b/tests/test_litellm/llms/azure/test_audio_transcriptions.py index 4f1906d80be..267ddaf9a5d 100644 --- a/tests/test_litellm/llms/azure/test_audio_transcriptions.py +++ b/tests/test_litellm/llms/azure/test_audio_transcriptions.py @@ -1,6 +1,8 @@ +import io import json from pathlib import Path from typing import Final +from unittest.mock import MagicMock import httpx import pytest @@ -9,6 +11,8 @@ from openai import AzureOpenAI import litellm from litellm.cost_calculator import completion_cost from litellm.litellm_core_utils.audio_utils.utils import calculate_request_duration +from litellm.llms.azure.audio_transcriptions import AzureAudioTranscription +from litellm.types.utils import TranscriptionResponse AUDIO_FILE: Final = Path(__file__).parents[3] / "gettysburg.wav" WHISPER_COST_PER_SECOND: Final = 0.0001 @@ -39,3 +43,43 @@ def test_azure_transcription_keeps_the_azure_provider(): assert response._hidden_params["custom_llm_provider"] == "azure" assert json.loads(response.model_dump_json())["text"] == "Four score and seven years ago" + + +@pytest.mark.parametrize("api_version", ["v1", "latest", "preview"]) +@pytest.mark.parametrize( + ("model", "expected_path"), + [ + ("whisper-1", "/openai/v1/audio/transcriptions"), + ("gpt-transcribe", "/openai/deployments/gpt-transcribe/audio/transcriptions"), + ("custom-transcribe-deployment", "/openai/deployments/custom-transcribe-deployment/audio/transcriptions"), + ], +) +def test_azure_transcription_alias_uses_model_route( + monkeypatch: pytest.MonkeyPatch, model: str, expected_path: str, api_version: str +) -> None: + def send_response(request: httpx.Request) -> httpx.Response: + assert request.url.path == expected_path + assert request.url.params.get("api-version") == ( + None if expected_path.startswith("/openai/v1/") else litellm.AZURE_DEFAULT_API_VERSION + ) + return httpx.Response(200, json={"text": "hello"}) + + audio_file: Final = io.BytesIO(b"audio") + audio_file.name = "sample.wav" + with httpx.Client(transport=httpx.MockTransport(send_response)) as http_client: + monkeypatch.setattr(litellm, "client_session", http_client) + monkeypatch.setattr(litellm, "model_cost", litellm.get_model_cost_map(url="")) + response: Final = AzureAudioTranscription().audio_transcriptions( + model=model, + audio_file=audio_file, + optional_params={"response_format": "json"}, + logging_obj=MagicMock(), + model_response=TranscriptionResponse(), + timeout=10, + max_retries=0, + api_key="test-key", + api_base="https://example.openai.azure.com", + api_version=api_version, + ) + + assert response.text == "hello" diff --git a/tests/unit/llms/openai/transcriptions/test_gpt_transcribe.py b/tests/unit/llms/openai/transcriptions/test_gpt_transcribe.py index d192550943c..21660fca9f9 100644 --- a/tests/unit/llms/openai/transcriptions/test_gpt_transcribe.py +++ b/tests/unit/llms/openai/transcriptions/test_gpt_transcribe.py @@ -251,40 +251,6 @@ def test_gpt_live_transcribe_rejects_file_transcription(local_model_cost_map: No ) -@pytest.mark.parametrize( - ("api_version", "expected_api_version"), - [ - ("v1", litellm.AZURE_DEFAULT_API_VERSION), - ("latest", litellm.AZURE_DEFAULT_API_VERSION), - ("preview", litellm.AZURE_DEFAULT_API_VERSION), - (None, None), - ("2025-04-01-preview", "2025-04-01-preview"), - ], -) -@pytest.mark.parametrize("model", ["gpt-transcribe", "custom-transcribe-deployment"]) -def test_azure_audio_transcription_resolves_api_version_in_provider( - model: str, api_version: str | None, expected_api_version: str | None -) -> None: - handler = AzureAudioTranscription() - handler.async_audio_transcriptions = MagicMock(return_value=MagicMock()) - - handler.audio_transcriptions( - model=model, - audio_file=io.BytesIO(b"audio"), - optional_params={"stream": True}, - logging_obj=MagicMock(), - model_response=TranscriptionResponse(), - timeout=10, - max_retries=0, - api_key="sk-test", - api_base="https://example.openai.azure.com", - api_version=api_version, - atranscription=True, - ) - - assert handler.async_audio_transcriptions.call_args.kwargs["api_version"] == expected_api_version - - def test_azure_gpt_transcribe_uses_deployment_scoped_route(): def send_response(request: httpx.Request) -> httpx.Response: assert str(request.url) == (