Repository navigation
Cache control injection uses wrong detection for Anthropic SDK models #14642
Description
Description
The applyCaching function in transform.ts uses model.providerID === "anthropic" to determine whether to use message-level cache control options. This is incorrect for non-Claude models (e.g. kimi-k2.5) that use the Anthropic SDK but have a different providerID.
Current behavior
const useMessageLevelOptions = model.providerID === "anthropic" || model.providerID.includes("bedrock")For a model configured as:
{
providerID: "kimi",
api: { npm: "@ai-sdk/anthropic", ... }
}Cache control is incorrectly placed at the content part level instead of the message level.
Expected behavior
Detection should be based on model.api.npm to correctly identify Anthropic SDK usage:
const isAnthropicSdk = model.api.npm === "@ai-sdk/anthropic" || model.api.npm === "@ai-sdk/google-vertex/anthropic"
const useMessageLevelOptions = isAnthropicSdk || model.providerID.includes("bedrock")Impact
Non-Claude models using @ai-sdk/anthropic may not have cache control applied correctly.
Activity
github-actions commentedon Feb 22, 2026
This issue might be a duplicate of existing issues. Please check:
applyCachingintransform.tsuses incorrect detection logic, causing cache control to be applied to models using non-Anthropic SDKs (or vice versa). That issue also proposes switching tomodel.api.npm-based detection as a fix.@ai-sdk/anthropic.If your issue is distinct from these, please clarify how it differs so it can be tracked separately.
fix: empty tool-result content and cache control level for custom @ai…
github-actions commentedon Apr 24, 2026
To stay organized issues are automatically closed after 60 days of no activity. If the issue is still relevant please open a new one.