fix(pricing): drop the unpublished cached rate from the Gemini Live preview entries

litellm/model_prices_and_context_window_backup.json

@@ -25882,7 +25882,6 @@
         "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",
@@ -25911,7 +25910,7 @@
         "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,
@@ -25927,7 +25926,6 @@
         "input_cost_per_image_token": 3e-06
     },
     "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",
@@ -25957,7 +25955,7 @@
         "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,

model_prices_and_context_window.json

@@ -25882,7 +25882,6 @@
         "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",
@@ -25911,7 +25910,7 @@
         "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,
@@ -25927,7 +25926,6 @@
         "input_cost_per_image_token": 3e-06
     },
     "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",
@@ -25957,7 +25955,7 @@
         "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,

tests/integration/contracts.json

@@ -290,6 +290,12 @@
     "tests/integration/spend/test_spend_calculate.py::test_spend_calculate_rejects_unpriced_model_with_400": [
       "quota_management.spend_tracking.spend_calculate.rejects_unpriced_model"
     ],
+    "tests/integration/spend/test_spend_calculate.py::test_live_preview_entry_charges_cached_tokens_at_the_fresh_rate[gemini-live-2.5-flash-preview-native-audio-09-2025]": [
+      "quota_management.spend_tracking.spend_calculate.live_preview_cached_tokens_cost_fresh_rate"
+    ],
+    "tests/integration/spend/test_spend_calculate.py::test_live_preview_entry_charges_cached_tokens_at_the_fresh_rate[gemini/gemini-live-2.5-flash-preview-native-audio-09-2025]": [
+      "quota_management.spend_tracking.spend_calculate.live_preview_cached_tokens_cost_fresh_rate"
+    ],
     "tests/integration/management/test_partial_update_sequences.py::test_restricted_actor_cannot_detach_key_from_project": [
       "mgmt.key.update.project_detach_denied_to_restricted_actor"
     ],

tests/integration/spend/test_spend_calculate.py

@@ -18,3 +18,51 @@ def test_spend_calculate_rejects_unpriced_model_with_400(gateway: Gateway) -> No
     assert error["type"] == "invalid_request_error", response.text
     assert error["param"] == "model", response.text
     assert model in string_value(error["message"]), response.text
+
+
+GEMINI_LIVE_PREVIEW_MODELS: Final = (
+    "gemini-live-2.5-flash-preview-native-audio-09-2025",
+    "gemini/gemini-live-2.5-flash-preview-native-audio-09-2025",
+)
+
+
+@pytest.mark.parametrize("model", GEMINI_LIVE_PREVIEW_MODELS)
+@pytest.mark.covers("quota_management.spend_tracking.spend_calculate.live_preview_cached_tokens_cost_fresh_rate")
+def test_live_preview_entry_charges_cached_tokens_at_the_fresh_rate(gateway: Gateway, model: str) -> None:
+    def cost_with_cached_tokens(cached_tokens: int) -> float:
+        response: Final = gateway.request(
+            "POST",
+            "/spend/calculate",
+            {
+                "completion_response": {
+                    "id": "chatcmpl-live-preview",
+                    "object": "chat.completion",
+                    "created": 1677652288,
+                    "model": model,
+                    "choices": [
+                        {
+                            "index": 0,
+                            "message": {"role": "assistant", "content": "live preview answer"},
+                            "finish_reason": "stop",
+                        }
+                    ],
+                    "usage": {
+                        "prompt_tokens": 101_000,
+                        "completion_tokens": 0,
+                        "total_tokens": 101_000,
+                        "prompt_tokens_details": {"cached_tokens": cached_tokens},
+                    },
+                }
+            },
+        )
+        assert response.status_code == 200, response.text
+        cost: Final = object_value(JSON_OBJECT.validate_json(response.text))["cost"]
+        assert isinstance(cost, int | float)
+        return float(cost)
+
+    cached_cost: Final = cost_with_cached_tokens(100_000)
+    fresh_cost: Final = cost_with_cached_tokens(0)
+    assert fresh_cost > 0, fresh_cost
+    assert cached_cost == pytest.approx(fresh_cost), (
+        f"the entry publishes no cached rate, so 100k cached tokens must bill like fresh ones: {cached_cost} vs {fresh_cost}"
+    )

tests/test_litellm/test_cost_calculator.py

@@ -38,6 +38,7 @@
     Usage,
 )
 from litellm.types.videos.main import VideoObject
+from litellm.utils import supports_prompt_caching
 
 
 @pytest.fixture
@@ -4502,3 +4503,99 @@ def test_cost_per_token_bedrock_nemotron_super_3_uses_eu_west_2_entry_not_us_rat
 
     assert prompt_usd == pytest.approx(prompt_tokens * regional["input_cost_per_token"])
     assert completion_usd == pytest.approx(completion_tokens * regional["output_cost_per_token"])
+
+
+GPT_REALTIME_2_FAMILY: Final = (
+    "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:
+    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
+    }
+
+    # Azure publishes one cached-audio meter per gpt-realtime-2 deployment,
+    # https://azure.microsoft.com/en-us/pricing/details/cognitive-services/openai-service/, checked 2026-09-23
+    assert all(read is not None and write == read for read, write in audio_cache_rates.values()), audio_cache_rates
+    assert len(audio_cache_rates) == len(GPT_REALTIME_2_FAMILY)
+
+
+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:
+    # the Vertex pricing table prints N/A for cached input on every Live row,
+    # https://cloud.google.com/vertex-ai/generative-ai/pricing, checked 2026-09-23
+    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),
+        ),
+    )
+    fresh_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),
+    )
+
+    assert prompt_usd == pytest.approx(fresh_usd), (
+        "with no cached rate the cached tokens bill at the input rate, so a phantom discount cannot appear"
+    )
+    assert prompt_usd > 0
+
+
+@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 Vertex context-caching supported-model lists contain no Live model while 2.5 Flash is listed,
+    # https://cloud.google.com/vertex-ai/generative-ai/docs/context-cache/context-cache-overview, checked 2026-09-23
+    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"
+    )
+
+
+@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:
+    info = litellm.get_model_info(model)
+
+    # the Vertex model card for gemini-live-2.5-flash-native-audio publishes these limits and flags,
+    # https://cloud.google.com/vertex-ai/generative-ai/docs/models, checked 2026-09-23
+    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