diff --git a/packages/core/src/session/runner/to-llm-message.ts b/packages/core/src/session/runner/to-llm-message.ts
index 51b0f1ef1ef..9c90d775f60 100644
--- a/packages/core/src/session/runner/to-llm-message.ts
+++ b/packages/core/src/session/runner/to-llm-message.ts
@@ -157,6 +157,7 @@ const assistant = (message: SessionMessage.Assistant, model: Model.Ref, provider
           providerMetadata: reuseProviderMetadata ? providerMetadata(providerMetadataKey, item.state) : undefined,
         },
       ]
+    // Let the destination adapter handle readable reasoning after a model/provider switch.
     if (item.type === "reasoning")
       return reuseProviderMetadata
         ? [
@@ -167,7 +168,7 @@ const assistant = (message: SessionMessage.Assistant, model: Model.Ref, provider
             },
           ]
         : item.text.length > 0
-          ? [{ type: "text", text: item.text }]
+          ? [{ type: message.error === undefined ? "reasoning" : "text", text: item.text }]
           : []
     // Call-side metadata is model-scoped proof of generation (Gemini thought
     // signatures, OpenAI encrypted reasoning): only the producing model may
diff --git a/packages/core/test/session-runner-message.test.ts b/packages/core/test/session-runner-message.test.ts
index 706ecc30227..462e3f6c919 100644
--- a/packages/core/test/session-runner-message.test.ts
+++ b/packages/core/test/session-runner-message.test.ts
@@ -984,7 +984,7 @@ Recent work
     )
 
     expect(messages[0]?.content).toEqual([
-      { type: "text", text: "Visible thought" },
+      { type: "reasoning", text: "Visible thought" },
       {
         type: "tool-call",
         id: "hosted-old-model",