Bedrock prompt caching is not working

[LokeshEka]

Initial Checks

• I'm using the latest version of Pydantic AI
• I've searched for my issue in the issue tracker before opening this issue

Description

I am using inference profile arn as model id

@DouweM Am I doing anything wrong here?

Minimal, Reproducible Example

data_agent = Agent(
    BedrockConverseModel(AGENT_SETTINGS.model),
    deps_type=AgentDeps,
    retries=AGENT_SETTINGS.retries,
    instructions=SYSTEM_INSTRUCTIONS,
    tools=[execute_sql, search_tables, clarify_with_user],
    model_settings=BedrockModelSettings(
        bedrock_cache_instructions=True,  # Cache system instructions
        bedrock_cache_tool_definitions=True,  # Cache tool definitions
        bedrock_cache_messages=True,  # Also cache the last message
    ),
)

print(result.usage())
RunUsage(input_tokens=44334, output_tokens=1279, requests=5, tool_calls=4)

Logfire Trace

No response

Python, Pydantic AI & LLM client version

• Python: 3.13.11
• Pydantic AI: 1.59.0
• LLM provider SDK: Bedrock

[DouweM]

@LokeshEka Can you try to capture what's being sent to Bedrock using https://ai.pydantic.dev/logfire/#monitoring-http-requests to verify if the cache points are making it into the request payload?

Also note that not all models on Bedrock support prompt caching, so it's very relevant to know which model you're using.

[LokeshEka]

Hi @DouweM http request for bedrock are not traced in logfire but agent tool's http requests are traced

I am using sonnet 4.6 inference profile arn

From my understanding, caching support is now hardcoded based on the model name:

pydantic-ai/pydantic_ai_slim/pydantic_ai/providers/bedrock.py

Line 112 in bc7f6a2

'anthropic': lambda model_name: replace(

so if we are using inference profile ARN prompt caching is disabled

[LokeshEka]

As the docs say, this will solve the issue

# Create provider with your AWS configuration
provider = BedrockProvider(region_name="ap-south-1")

# Create a profile with the logical model name for feature detection
profile = provider.model_profile("model name) # which has anthropic in it so that it will be used for caching