Does Bedrock include outputConfig in its prompt caching key?

Question

Hi there. When using Bedrock's Converse API with both cachePoint in the system prompt and outputConfig for structured output, it looks like only changing the schema between requests causes a prompt caching miss. I could not see this being documented anywhere. I found this to be an unexpected interaction between prompt caching and structured output that should either be documented or fixed.

Ruby illustration below:
```
require "aws-sdk-bedrockruntime"
require "json"
require "faker"

client = Aws::BedrockRuntime::Client.new(region: "us-west-2")

TEXT = Array.new(60) { Faker::Lorem.paragraph(sentence_count: 5) }.join("

")

SYSTEM_PROMPT = <<~PROMPT
Summarize the text below:

PROMPT

SCHEMA_A = JSON.generate({
type: "object",
properties: { summary: { type: "string" }, key_points: { type: "array", items: { type: "string" } } },
required: ["summary", "key_points"],
additionalProperties: false
})

SCHEMA_B = JSON.generate({
type: "object",
properties: { topic: { type: "string" } },
required: ["topic"],
additionalProperties: false
})

def call(client, label, schema)
resp = client.converse(
    model_id: "us.anthropic.claude-sonnet-4-6",
    system: [
    { text: SYSTEM_PROMPT },
    { cache_point: { type: "default" } }
    ],
    messages: [{ role: "user", content: [{ text: "Summarize this text." }] }],
    inference_config: { temperature: 0 },
    output_config: {
    text_format: {
        type: "json_schema",
        structure: {
        json_schema: { schema: schema, name: label }
        }
    }
    }
)
u = resp.usage
puts "#{label}: input=#{u.input_tokens} cached_hit=#{u.cache_read_input_tokens}
cache_created=#{u.cache_write_input_tokens}"
end

puts "
1. Warm up cache with Schema A"
call(client, "schema_a", SCHEMA_A)

puts "2. Same schema — expect a cache HIT"
call(client, "schema_a", SCHEMA_A)

puts "3. Different schema — expect a cache HIT but it is MISS"
call(client, "schema_b", SCHEMA_B)

puts "4. Back to Schema A — expect a cache HIT"
call(client, "schema_a", SCHEMA_A)
```

Florian Turnwald · Answer

Your observation is spot on. While not explicitly detailed in the docs, it’s a known behavior in managed LLM environments: outputConfig (and specifically the JSON schema) is treated as part of the model's instruction state.
Since the schema dictates how the model must structure its response, it is bundled into the cache key. Consequently, even if your system prompt remains identical, a change in the schema results in a different hash and a cache miss. To get a hit, the entire request configuration up to (and including) the cache_point must remain static.

re:Post Agent · Answer

Based on the available documentation, I cannot find explicit information about whether `outputConfig` is included in Bedrock's prompt caching key. The documentation describes `outputConfig` as an optional parameter that controls the output configuration for a model response in calls to Converse or ConverseStream, and it includes a `textFormat` property for structured output parameters. However, the interaction between `outputConfig` (specifically JSON schema definitions) and prompt caching is not documented in the sources provided.

Your observation that changing the JSON schema in `outputConfig` causes a prompt cache miss is an important finding. This suggests that the schema definition may indeed be part of the caching key, even though the cached content (the system prompt with cache point) remains unchanged. This behavior would mean that different structured output schemas are treated as distinct requests for caching purposes, which could have significant implications for cache efficiency when using structured outputs with varying schemas.

This appears to be an undocumented aspect of how prompt caching interacts with structured outputs. Given that this behavior affects cache hit rates and could impact both performance and costs, it would be valuable to have this interaction explicitly documented or clarified in the official documentation.  
**Sources**  
[OutputConfig - Amazon Bedrock](https://docs.aws.amazon.com/bedrock/latest/APIReference/API_runtime_OutputConfig.html)  
[Class: Aws::BedrockRuntime::Types::OutputConfig
  
    — AWS SDK for Ruby V3](https://docs.aws.amazon.com/sdk-for-ruby/v3/api/Aws/BedrockRuntime/Types/OutputConfig.html)  
[Converse - Amazon Bedrock](https://docs.aws.amazon.com/bedrock/latest/APIReference/API_runtime_Converse.html)

Does Bedrock include outputConfig in its prompt caching key?

Add your answer

Relevant content