bug: aws bedrock prompt caching doesn't work correctly

[rgodha24]

Description

prompt caching on aws bedrock doesnt work

this makes running opencode thru Opus on AWS Bedrock like 10x more expensive than running it on Anthropic API.

AI diagnosis:

Evidence

Before fix:

Turn	Input	Cache Read	Cache Write
1	~11k	~13k	~11k
2	~11k	~13k	0

Plugins

n/a

OpenCode version

1.1.48

Steps to reproduce

use any claude model on aws bedrock in opencode

only the system prompt (e.g. the first ~13k tokens) is read from cache

this means that its like 10x more expensive than running opencode thru official anthropic API

Screenshot and/or share link

No response

Operating System

No response

Terminal

No response

[github-actions]

This issue might be a duplicate of existing issues. Please check:

• Excessive input token usage with Sonnet on AWS Bedrock #10066: Excessive input token usage with Sonnet on AWS Bedrock (describes the same 10x cost increase and indicates tool outputs are being repeatedly included)
• Claude w/ bedrock custom inference profile - no caching #9803: Claude w/ bedrock custom inference profile - no caching (directly addresses bedrock prompt caching detection issues)
• Due to the lack of userId and sessionId, some models will not hit the cache. #8623: Due to the lack of userId and sessionId, some models will not hit the cache (related to cache metadata)
• The Claude model cannot properly enable the caching function #11083: The Claude model cannot properly enable the caching function (related to caching detection with Claude)

Feel free to ignore if none of these address your specific case.

[thevinchi]

This has been working great, but the same behavior is back when attempting to use Opus 4.6 on bedrock. I assume I should open a new issue?

[thevinchi]

Nevermind, I had not updated opencode to the latest version in the environment where we were testing Opus 4.6. 🤦‍♂️