diff --git a/litellm/model_prices_and_context_window_backup.json b/litellm/model_prices_and_context_window_backup.json index 7fa09951eae..79a915cf5ac 100644 --- a/litellm/model_prices_and_context_window_backup.json +++ b/litellm/model_prices_and_context_window_backup.json @@ -5578,6 +5578,7 @@ "supports_tool_choice": true }, "azure/gpt-realtime-2": { + "cache_creation_input_audio_token_cost": 4e-07, "cache_read_input_audio_token_cost": 4e-07, "cache_read_input_token_cost": 4e-07, "deprecation_date": "2026-08-31", @@ -24152,7 +24153,6 @@ "input_cost_per_image_token": 3e-06 }, "gemini-live-2.5-flash-preview-native-audio-09-2025": { - "cache_read_input_token_cost": 7.5e-08, "input_cost_per_audio_token": 3e-06, "input_cost_per_token": 5e-07, "litellm_provider": "vertex_ai-language-models", @@ -24181,7 +24181,7 @@ "supports_function_calling": true, "supports_parallel_function_calling": true, "supports_pdf_input": true, - "supports_prompt_caching": true, + "supports_prompt_caching": false, "supports_response_schema": true, "supports_system_messages": true, "supports_tool_choice": true, @@ -24196,7 +24196,6 @@ "gemini_native_audio": true }, "gemini/gemini-live-2.5-flash-preview-native-audio-09-2025": { - "cache_read_input_token_cost": 7.5e-08, "input_cost_per_audio_token": 3e-06, "input_cost_per_token": 5e-07, "litellm_provider": "gemini", @@ -24226,7 +24225,7 @@ "supports_function_calling": true, "supports_parallel_function_calling": true, "supports_pdf_input": true, - "supports_prompt_caching": true, + "supports_prompt_caching": false, "supports_response_schema": true, "supports_system_messages": true, "supports_tool_choice": true, diff --git a/model_prices_and_context_window.json b/model_prices_and_context_window.json index 7fa09951eae..79a915cf5ac 100644 --- a/model_prices_and_context_window.json +++ b/model_prices_and_context_window.json @@ -5578,6 +5578,7 @@ "supports_tool_choice": true }, "azure/gpt-realtime-2": { + "cache_creation_input_audio_token_cost": 4e-07, "cache_read_input_audio_token_cost": 4e-07, "cache_read_input_token_cost": 4e-07, "deprecation_date": "2026-08-31", @@ -24152,7 +24153,6 @@ "input_cost_per_image_token": 3e-06 }, "gemini-live-2.5-flash-preview-native-audio-09-2025": { - "cache_read_input_token_cost": 7.5e-08, "input_cost_per_audio_token": 3e-06, "input_cost_per_token": 5e-07, "litellm_provider": "vertex_ai-language-models", @@ -24181,7 +24181,7 @@ "supports_function_calling": true, "supports_parallel_function_calling": true, "supports_pdf_input": true, - "supports_prompt_caching": true, + "supports_prompt_caching": false, "supports_response_schema": true, "supports_system_messages": true, "supports_tool_choice": true, @@ -24196,7 +24196,6 @@ "gemini_native_audio": true }, "gemini/gemini-live-2.5-flash-preview-native-audio-09-2025": { - "cache_read_input_token_cost": 7.5e-08, "input_cost_per_audio_token": 3e-06, "input_cost_per_token": 5e-07, "litellm_provider": "gemini", @@ -24226,7 +24225,7 @@ "supports_function_calling": true, "supports_parallel_function_calling": true, "supports_pdf_input": true, - "supports_prompt_caching": true, + "supports_prompt_caching": false, "supports_response_schema": true, "supports_system_messages": true, "supports_tool_choice": true, diff --git a/tests/test_litellm/test_cost_calculator.py b/tests/test_litellm/test_cost_calculator.py index f610821e06a..3c857c59cf2 100644 --- a/tests/test_litellm/test_cost_calculator.py +++ b/tests/test_litellm/test_cost_calculator.py @@ -26,7 +26,7 @@ from litellm.types.utils import ( PromptTokensDetailsWrapper, Usage, ) -from litellm.utils import TranscriptionResponse +from litellm.utils import TranscriptionResponse, supports_prompt_caching @pytest.fixture @@ -305,6 +305,71 @@ def test_github_copilot_mai_code_1_flash_pricing(_local_model_cost_map, model): assert completion_usd == pytest.approx(500 * 4.5e-06) +GPT_REALTIME_2_FAMILY: Final = ( + "azure/gpt-realtime-2", + "azure/gpt-realtime-2.1", + "azure/gpt-realtime-2.1-mini", + "gpt-realtime-2", + "gpt-realtime-2.1", + "gpt-realtime-2.1-mini", +) + + +def test_gpt_realtime_2_family_prices_audio_cache_writes_and_reads_alike(_local_model_cost_map: None) -> None: + """Azure publishes one cached-audio meter per gpt-realtime-2 deployment, so the write rate equals the read.""" + audio_cache_rates: Final = { + model: ( + litellm.model_cost[model].get("cache_read_input_audio_token_cost"), + litellm.model_cost[model].get("cache_creation_input_audio_token_cost"), + ) + for model in GPT_REALTIME_2_FAMILY + } + + assert all(read is not None and write == read for read, write in audio_cache_rates.values()), audio_cache_rates + assert audio_cache_rates["azure/gpt-realtime-2"] == (4e-07, 4e-07) + + +GEMINI_LIVE_NATIVE_AUDIO_CASES: Final = ( + ("gemini-live-2.5-flash-native-audio", "vertex_ai"), + ("gemini-live-2.5-flash-preview-native-audio-09-2025", "vertex_ai"), + ("gemini/gemini-live-2.5-flash-preview-native-audio-09-2025", "gemini"), +) + + +@pytest.mark.parametrize(("model", "provider"), GEMINI_LIVE_NATIVE_AUDIO_CASES) +def test_gemini_live_native_audio_carries_no_cached_input_rate( + _local_model_cost_map: None, model: str, provider: str +) -> None: + """Google prints N/A in both cached columns for every Live API row, so no cached-input rate can be charged.""" + assert litellm.get_model_info(model, custom_llm_provider=provider)["cache_read_input_token_cost"] is None + + prompt_usd, _ = cost_per_token( + model=model, + prompt_tokens=101_000, + completion_tokens=0, + custom_llm_provider=provider, + usage_object=Usage( + prompt_tokens=101_000, + completion_tokens=0, + prompt_tokens_details=PromptTokensDetailsWrapper(cached_tokens=100_000), + ), + ) + + assert prompt_usd == pytest.approx(1_000 * 5e-07), "the 100k cached tokens drop out with no cached rate to charge" + + +@pytest.mark.parametrize(("model", "provider"), GEMINI_LIVE_NATIVE_AUDIO_CASES) +def test_gemini_live_native_audio_declares_prompt_caching_unsupported( + _local_model_cost_map: None, model: str, provider: str +) -> None: + """The vendor's documented no has to be recorded as False, since an absent key reads back as None.""" + assert litellm.get_model_info(model, custom_llm_provider=provider)["supports_prompt_caching"] is False + assert supports_prompt_caching(model=model, custom_llm_provider=provider) is False + assert supports_prompt_caching(model="gemini-2.5-flash", custom_llm_provider="vertex_ai") is True, ( + "control: the helper swallows a lookup error into False, so without this a broken lookup reads as a pass" + ) + + def test_cost_calculator_with_usage(_local_model_cost_map, monkeypatch): usage = Usage( @@ -4501,6 +4566,24 @@ def test_gemini_live_native_audio_ga_realtime_cost(_local_model_cost_map: None) assert cost == pytest.approx(expected_cost, rel=1e-9) +@pytest.mark.parametrize( + "model", + ["gemini-live-2.5-flash-native-audio", "vertex_ai/gemini-live-2.5-flash-native-audio"], +) +def test_gemini_live_native_audio_limits_and_capabilities_match_vendor_model_card( + _local_model_cost_map: None, model: str +) -> None: + """Google's card for model ID gemini-live-2.5-flash-native-audio is the source for these limits and flags.""" + info = litellm.get_model_info(model) + + assert info["max_input_tokens"] == 131072 + assert info["max_output_tokens"] == 65536 + assert info["max_tokens"] == 65536 + assert info["supports_response_schema"] is False + assert info["supports_url_context"] is False + assert info["supports_pdf_input"] is False + + @pytest.mark.parametrize( "priceless_entry", [