diff --git a/packages/runtime/src/__tests__/ai-sdk-backend.test.ts b/packages/runtime/src/__tests__/ai-sdk-backend.test.ts
index 97e7c4c76c..51901a2e81 100644
--- a/packages/runtime/src/__tests__/ai-sdk-backend.test.ts
+++ b/packages/runtime/src/__tests__/ai-sdk-backend.test.ts
@@ -79,6 +79,7 @@ import {
 import type { MemoryExtractionSourceSnapshot } from '../memory-extraction.js';
 import type { OpenAiResponsesSemanticBaseline } from '../openai-responses-continuation.js';
 import type { OpenAiResponsesTransportState } from '../openai-responses-websocket.js';
+import { getAIModel } from '../model-factory.js';
 
 describe('AiSdkBackend ApplyPatch routing', () => {
   test('advertises apply_patch only to supported native OpenAI models', async () => {
@@ -10809,6 +10810,178 @@ describe('AiSdkBackend thinking persistence', () => {
     assert.ok(prompt.indexOf('reasoning about the tool result') < prompt.indexOf('tool-1'));
   });
 
+  test('omits Responses reasoning without encrypted content from the wire request', async (t) => {
+    for (const replayCase of [
+      { name: 'missing', openai: { itemId: 'rs_deepseek' } },
+      {
+        name: 'null',
+        openai: { itemId: 'rs_deepseek', reasoningEncryptedContent: null },
+      },
+      {
+        name: 'empty string',
+        openai: { itemId: 'rs_deepseek', reasoningEncryptedContent: '' },
+      },
+    ] as const) {
+      await t.test(replayCase.name, async () => {
+        const runtimeContext: RuntimeEvent[] = [
+          runtimeEvent({
+            id: 'e1',
+            turnId: 'turn-prev',
+            role: 'model',
+            author: 'agent',
+            content: {
+              kind: 'thinking',
+              text: 'plaintext reasoning from DeepSeek',
+              providerOptions: { openai: replayCase.openai },
+            },
+            refs: { providerEventId: 'm1' },
+          }),
+          runtimeEvent({
+            id: 'e2',
+            turnId: 'turn-prev',
+            role: 'model',
+            author: 'agent',
+            content: { kind: 'text', text: 'answer before the tool call' },
+            refs: { providerEventId: 'm1' },
+          }),
+          runtimeEvent({
+            id: 'e3',
+            turnId: 'turn-prev',
+            role: 'model',
+            author: 'agent',
+            content: {
+              kind: 'function_call',
+              id: 'tool-1',
+              name: 'Read',
+              args: { path: 'package.json' },
+            },
+            refs: { toolCallId: 'tool-1', stepId: 'm1' },
+          }),
+          runtimeEvent({
+            id: 'e4',
+            turnId: 'turn-prev',
+            role: 'tool',
+            author: 'tool',
+            content: {
+              kind: 'function_response',
+              id: 'tool-1',
+              name: 'Read',
+              result: { kind: 'text', text: 'file contents' },
+            },
+            refs: { toolCallId: 'tool-1' },
+          }),
+        ];
+        let requestBody: Record<string, unknown> | undefined;
+        const fetch = (async (_url: string | URL | Request, init?: RequestInit) => {
+          requestBody = JSON.parse(String(init?.body)) as Record<string, unknown>;
+          const events = [
+            { type: 'response.created', response: { id: 'response-current' } },
+            {
+              type: 'response.completed',
+              response: {
+                id: 'response-current',
+                object: 'response',
+                created_at: 8,
+                model: 'deepseek-v4-flash',
+                status: 'completed',
+                output: [],
+                usage: { input_tokens: 1, output_tokens: 1 },
+              },
+            },
+          ];
+          const body = `${events
+            .map((event) => `data: ${JSON.stringify(event)}`)
+            .join('\n\n')}\n\ndata: [DONE]\n\n`;
+          return new Response(body, {
+            status: 200,
+            headers: { 'content-type': 'text/event-stream' },
+          });
+        }) as unknown as typeof globalThis.fetch;
+        const secondBackend = createTestAiSdkBackend({
+          sessionId: 'session-1',
+          header: header(),
+          appendMessage: async () => {},
+          connection: {
+            slug: 'deepseek',
+            providerType: 'deepseek',
+            defaultModel: 'deepseek-v4-flash',
+          },
+          apiKey: 'deepseek-test-token',
+          modelId: 'deepseek-v4-flash',
+          modelFactory: (input) => getAIModel({ ...input, fetch }),
+          tools: [],
+          newId: idGenerator(),
+          now: monotonicClock(),
+        });
+
+        await drain(
+          secondBackend.send({
+            turnId: 'turn-current',
+            text: 'follow up',
+            context: [],
+            runtimeContext,
+          }),
+        );
+
+        const input = requestBody?.input;
+        assert.ok(Array.isArray(input));
+        assert.equal(
+          input.some((item) => item?.type === 'reasoning'),
+          false,
+        );
+        assert.deepEqual(
+          input.slice(0, 3).map((item) => {
+            assert.ok(item && typeof item === 'object' && !Array.isArray(item));
+            const record = item as Record<string, unknown>;
+            const content = Array.isArray(record.content) ? record.content : [];
+            const firstContent = content[0];
+            return {
+              type: record.type ?? (typeof record.role === 'string' ? 'message' : undefined),
+              role: record.role,
+              text:
+                firstContent && typeof firstContent === 'object' && !Array.isArray(firstContent)
+                  ? (firstContent as Record<string, unknown>).text
+                  : undefined,
+              callId: record.call_id,
+              name: record.name,
+              arguments: record.arguments,
+              output: record.output,
+            };
+          }),
+          [
+            {
+              type: 'message',
+              role: 'assistant',
+              text: 'answer before the tool call',
+              callId: undefined,
+              name: undefined,
+              arguments: undefined,
+              output: undefined,
+            },
+            {
+              type: 'function_call',
+              role: undefined,
+              text: undefined,
+              callId: 'tool-1',
+              name: 'Read',
+              arguments: '{"path":"package.json"}',
+              output: undefined,
+            },
+            {
+              type: 'function_call_output',
+              role: undefined,
+              text: undefined,
+              callId: 'tool-1',
+              name: undefined,
+              arguments: undefined,
+              output: '{"kind":"text","text":"file contents"}',
+            },
+          ],
+        );
+      });
+    }
+  });
+
   test('OpenAI Responses reasoning from a tool step is replayed with its encrypted content', async () => {
     const ctx = {
       sessionId: 'session-1',
diff --git a/packages/runtime/src/__tests__/model-adapter.test.ts b/packages/runtime/src/__tests__/model-adapter.test.ts
index 04dc18dd0d..f7dd926252 100644
--- a/packages/runtime/src/__tests__/model-adapter.test.ts
+++ b/packages/runtime/src/__tests__/model-adapter.test.ts
@@ -67,7 +67,7 @@ describe('ModelAdapter stream and error normalization', () => {
       toolResults: true,
       signedThinking: false,
       unsignedThinking: true,
-      openAiResponsesThinking: false,
+      openAiResponsesEncryptedThinking: false,
     });
   });
 
@@ -121,7 +121,7 @@ describe('ModelAdapter stream and error normalization', () => {
       toolResults: true,
       signedThinking: false,
       unsignedThinking: false,
-      openAiResponsesThinking: true,
+      openAiResponsesEncryptedThinking: true,
     });
   });
 
diff --git a/packages/runtime/src/ai-sdk-backend.ts b/packages/runtime/src/ai-sdk-backend.ts
index ef72a73ad7..b9293f0834 100644
--- a/packages/runtime/src/ai-sdk-backend.ts
+++ b/packages/runtime/src/ai-sdk-backend.ts
@@ -3752,14 +3752,19 @@ export class AiSdkBackend implements AgentBackend {
             }
           : undefined;
       }
-      if (replaySupport.openAiResponsesThinking) {
+      if (replaySupport.openAiResponsesEncryptedThinking) {
         const openai = item.providerOptions?.openai;
         if (openai && typeof openai === 'object' && !Array.isArray(openai)) {
           const { itemId, reasoningEncryptedContent } = openai as {
             itemId?: unknown;
             reasoningEncryptedContent?: unknown;
           };
-          if (typeof itemId === 'string' && itemId.length > 0) {
+          if (
+            typeof itemId === 'string' &&
+            itemId.length > 0 &&
+            typeof reasoningEncryptedContent === 'string' &&
+            reasoningEncryptedContent.length > 0
+          ) {
             return {
               part: {
                 type: 'reasoning' as const,
@@ -3767,10 +3772,7 @@ export class AiSdkBackend implements AgentBackend {
                 providerOptions: {
                   openai: {
                     itemId,
-                    ...(typeof reasoningEncryptedContent === 'string' ||
-                    reasoningEncryptedContent === null
-                      ? { reasoningEncryptedContent }
-                      : {}),
+                    reasoningEncryptedContent,
                   },
                 },
               },
diff --git a/packages/runtime/src/model-adapter.ts b/packages/runtime/src/model-adapter.ts
index eee32bfdea..01b5782f7f 100644
--- a/packages/runtime/src/model-adapter.ts
+++ b/packages/runtime/src/model-adapter.ts
@@ -185,7 +185,8 @@ export class ModelAdapter {
       // relays that don't need the field ignore it. Reasoning is still
       // recorded to the event log and rendered regardless.
       unsignedThinking: this.runtime.reasoningReplay.kind === 'openai-chat-plaintext',
-      openAiResponsesThinking: this.runtime.reasoningReplay.kind === 'openai-responses-item',
+      openAiResponsesEncryptedThinking:
+        this.runtime.reasoningReplay.kind === 'openai-responses-encrypted',
     };
   }
 
@@ -634,7 +635,7 @@ export interface ModelAdapterRuntimeEventReplaySupport {
   toolResults: boolean;
   signedThinking: boolean;
   unsignedThinking: boolean;
-  openAiResponsesThinking: boolean;
+  openAiResponsesEncryptedThinking: boolean;
 }
 
 /**
diff --git a/packages/runtime/src/model-runtime.ts b/packages/runtime/src/model-runtime.ts
index f33ff92eb4..0d49a2bc64 100644
--- a/packages/runtime/src/model-runtime.ts
+++ b/packages/runtime/src/model-runtime.ts
@@ -20,7 +20,7 @@ export type ReasoningReplayContract =
   | { kind: 'none' }
   | { kind: 'anthropic-signed' }
   | { kind: 'openai-chat-plaintext'; requestField: 'observed' | 'reasoning' }
-  | { kind: 'openai-responses-item' };
+  | { kind: 'openai-responses-encrypted' };
 
 export interface ResolvedModelRuntime {
   adapter: ProviderRuntimeAdapter;
@@ -161,7 +161,11 @@ function reasoningReplayContract(
     case 'anthropic-messages':
       return { kind: 'anthropic-signed' };
     case 'openai-responses':
-      return { kind: 'openai-responses-item' };
+      // The native OpenAI serializer can replay only provider-issued encrypted
+      // reasoning when store=false. Open Responses plaintext reasoning is read
+      // by a separate response transport, but has no request codec here yet;
+      // @ai-sdk/open-responses unlocks an open-responses-plaintext sibling.
+      return { kind: 'openai-responses-encrypted' };
     case 'openai-chat':
       return adapter.kind === 'openai-compatible'
         ? {