core: fix message caching for Anthropic models to improve response consistency

packages/opencode/src/provider/transform.ts

@@ -205,7 +205,12 @@ export namespace ProviderTransform {
   export function message(msgs: ModelMessage[], model: Provider.Model) {
     msgs = unsupportedParts(msgs, model)
     msgs = normalizeMessages(msgs, model)
-    if (model.providerID === "anthropic" || model.api.id.includes("anthropic") || model.api.id.includes("claude")) {
+    if (
+      model.providerID === "anthropic" ||
+      model.api.id.includes("anthropic") ||
+      model.api.id.includes("claude") ||
+      model.api.npm === "@ai-sdk/anthropic"
+    ) {
       msgs = applyCaching(msgs, model.providerID)
     }