Merge pull request #14620 from BerriAI/litellm_dev_09_16_2025_p2

(feat) Anthropic - document pricing for cache creation tokens above 1hr

.gitignore

@@ -95,4 +95,5 @@ test.py
 litellm_config.yaml
 .cursor
 .vscode/launch.json
-litellm/proxy/to_delete_loadtest_work/*
+litellm/proxy/to_delete_loadtest_work/*
+update_model_cost_map.py

litellm/llms/anthropic/chat/transformation.py

@@ -200,8 +200,12 @@ class AnthropicConfig(AnthropicModelInfo, BaseConfig):
             )
 
             _allowed_properties = set(AnthropicInputSchema.__annotations__.keys())
-            input_schema_filtered = {k: v for k, v in _input_schema.items() if k in _allowed_properties}
-            input_anthropic_schema: AnthropicInputSchema = AnthropicInputSchema(**input_schema_filtered)
+            input_schema_filtered = {
+                k: v for k, v in _input_schema.items() if k in _allowed_properties
+            }
+            input_anthropic_schema: AnthropicInputSchema = AnthropicInputSchema(
+                **input_schema_filtered
+            )
 
             _tool = AnthropicMessagesTool(
                 name=tool["function"]["name"],

litellm/types/utils.py

@@ -123,6 +123,7 @@ class ModelInfoBase(ProviderSpecificModelInfo, total=False):
     max_output_tokens: Required[Optional[int]]
     input_cost_per_token: Required[float]
     cache_creation_input_token_cost: Optional[float]
+    cache_creation_input_token_cost_above_1hr: Optional[float]
     cache_read_input_token_cost: Optional[float]
     input_cost_per_character: Optional[float]  # only for vertex ai models
     input_cost_per_audio_token: Optional[float]
@@ -162,7 +163,9 @@ class ModelInfoBase(ProviderSpecificModelInfo, total=False):
         SearchContextCostPerQuery
     ]  # Cost for using web search tool
     citation_cost_per_token: Optional[float]  # Cost per citation token for Perplexity
-    tiered_pricing: Optional[List[Dict[str, Any]]]  # Tiered pricing structure for models like Dashscope
+    tiered_pricing: Optional[
+        List[Dict[str, Any]]
+    ]  # Tiered pricing structure for models like Dashscope
     litellm_provider: Required[str]
     mode: Required[
         Literal[

tests/local_testing/test_anthropic_prompt_caching.py

@@ -300,6 +300,76 @@ async def test_anthropic_api_prompt_caching_basic():
     )
 
 
+@pytest.mark.asyncio()
+async def test_anthropic_api_prompt_caching_basic_with_cache_creation():
+    from uuid import uuid4
+
+    random_id = uuid4()
+
+    litellm.set_verbose = True
+    response = await litellm.acompletion(
+        model="anthropic/claude-3-5-sonnet-20240620",
+        messages=[
+            # System Message
+            {
+                "role": "system",
+                "content": [
+                    {
+                        "type": "text",
+                        "text": "Here is the full text of a complex legal agreement {}".format(
+                            random_id
+                        )
+                        * 400,
+                        "cache_control": {"type": "ephemeral"},
+                    }
+                ],
+            },
+            # marked for caching with the cache_control parameter, so that this checkpoint can read from the previous cache.
+            {
+                "role": "user",
+                "content": [
+                    {
+                        "type": "text",
+                        "text": "What are the key terms and conditions in this agreement?",
+                        "cache_control": {"type": "ephemeral"},
+                    }
+                ],
+            },
+            {
+                "role": "assistant",
+                "content": "Certainly! the key terms and conditions are the following: the contract is 1 year long for $10/mo",
+            },
+            # The final turn is marked with cache-control, for continuing in followups.
+            {
+                "role": "user",
+                "content": [
+                    {
+                        "type": "text",
+                        "text": "What are the key terms and conditions in this agreement?",
+                        "cache_control": {"type": "ephemeral"},
+                    }
+                ],
+            },
+        ],
+        temperature=0.2,
+        max_tokens=10,
+        extra_headers={
+            "anthropic-version": "2023-06-01",
+            "anthropic-beta": "prompt-caching-2024-07-31",
+        },
+    )
+
+    print("response=", response)
+
+    assert "cache_read_input_tokens" in response.usage
+    assert "cache_creation_input_tokens" in response.usage
+
+    # Assert either a cache entry was created or cache was read - changes depending on the anthropic api ttl
+    assert (response.usage.cache_read_input_tokens > 0) or (
+        response.usage.cache_creation_input_tokens > 0
+    )
+
+
 @pytest.mark.asyncio()
 async def test_anthropic_api_prompt_caching_with_content_str():
     system_message = [

tests/test_litellm/test_utils.py

@@ -508,6 +508,7 @@ def test_aaamodel_prices_and_context_window_json_is_valid():
                 "supports_computer_use": {"type": "boolean"},
                 "cache_creation_input_audio_token_cost": {"type": "number"},
                 "cache_creation_input_token_cost": {"type": "number"},
+                "cache_creation_input_token_cost_above_1hr": {"type": "number"},
                 "cache_creation_input_token_cost_above_200k_tokens": {"type": "number"},
                 "cache_read_input_token_cost": {"type": "number"},
                 "cache_read_input_token_cost_above_200k_tokens": {"type": "number"},
@@ -661,16 +662,16 @@ def test_aaamodel_prices_and_context_window_json_is_valid():
                                 "type": "array",
                                 "items": {"type": "number"},
                                 "minItems": 2,
-                                "maxItems": 2
+                                "maxItems": 2,
                             },
                             "input_cost_per_token": {"type": "number"},
                             "output_cost_per_token": {"type": "number"},
                             "cache_read_input_token_cost": {"type": "number"},
-                            "output_cost_per_reasoning_token": {"type": "number"}
+                            "output_cost_per_reasoning_token": {"type": "number"},
                         },
                         "required": ["range"],
-                        "additionalProperties": False
-                    }
+                        "additionalProperties": False,
+                    },
                 },
             },
             "additionalProperties": False,
@@ -843,7 +844,6 @@ for commitment in BEDROCK_COMMITMENTS:
 print("block_list", block_list)
 
 
-
 def test_supports_computer_use_utility():
     """
     Tests the litellm.utils.supports_computer_use utility function.
@@ -925,8 +925,7 @@ def test_pre_process_non_default_params(model, custom_llm_provider):
     from litellm.utils import ProviderConfigManager, pre_process_non_default_params
 
     provider_config = ProviderConfigManager.get_provider_chat_config(
-        model=model, 
-        provider=LlmProviders(custom_llm_provider)
+        model=model, provider=LlmProviders(custom_llm_provider)
     )
 
     class ResponseFormat(BaseModel):