diff --git a/.changeset/normalize-responses-rate-limit.md b/.changeset/normalize-responses-rate-limit.md
new file mode 100644
index 0000000000..1ff4b7f167
--- /dev/null
+++ b/.changeset/normalize-responses-rate-limit.md
@@ -0,0 +1,6 @@
+---
+"@moonshot-ai/kosong": patch
+"@moonshot-ai/kimi-code": patch
+---
+
+Normalize malformed Responses stream rate limit errors as provider rate limit failures.
diff --git a/packages/kosong/src/providers/openai-responses.ts b/packages/kosong/src/providers/openai-responses.ts
index 59c0c852da..00e31e3cf9 100644
--- a/packages/kosong/src/providers/openai-responses.ts
+++ b/packages/kosong/src/providers/openai-responses.ts
@@ -1,5 +1,10 @@
 import type { ModelCapability } from '#/capability';
-import { APIContextOverflowError, ChatProviderError, isContextOverflowErrorCode } from '#/errors';
+import {
+  APIContextOverflowError,
+  APIStatusError,
+  ChatProviderError,
+  isContextOverflowErrorCode,
+} from '#/errors';
 import type { ContentPart, Message, StreamedMessagePart, ToolCall } from '#/message';
 import { extractText } from '#/message';
 import type {
@@ -111,6 +116,10 @@ function readStringField(object: RawObject, key: string): string | undefined {
   return typeof value === 'string' ? value : undefined;
 }
 
+function hasOwn(object: RawObject, key: string): boolean {
+  return Object.prototype.hasOwnProperty.call(object, key);
+}
+
 function readNullableStringField(object: RawObject, key: string): string | null | undefined {
   const value = object[key];
   if (value === null) return null;
@@ -237,9 +246,65 @@ function errorFromOpenAIResponsesEvent(
   if (isContextOverflowErrorCode(code)) {
     return new APIContextOverflowError(400, fullMessage);
   }
+  if (code === 'rate_limit_exceeded') {
+    return new APIStatusError(429, fullMessage);
+  }
   return new ChatProviderError(fullMessage);
 }
 
+function parseNestedGatewayStreamError(message: string):
+  | {
+      code: string | null;
+      message: string;
+      param: string | null;
+    }
+  | undefined {
+  const marker = 'received error while streaming:';
+  const markerIndex = message.indexOf(marker);
+  if (markerIndex === -1) return undefined;
+
+  const jsonText = message.slice(markerIndex + marker.length).trim();
+  if (jsonText.length === 0) return undefined;
+
+  let parsed: unknown;
+  try {
+    parsed = JSON.parse(jsonText);
+  } catch {
+    return undefined;
+  }
+
+  const error = asRawObject(parsed);
+  if (error === null) return undefined;
+
+  const nestedMessage = readStringField(error, 'message');
+  if (nestedMessage === undefined) return undefined;
+
+  return {
+    code: readNullableStringField(error, 'code') ?? null,
+    message: nestedMessage,
+    param: readNullableStringField(error, 'param') ?? null,
+  };
+}
+
+function malformedStreamErrorEvent(message: string): ChatProviderError {
+  const nested = parseNestedGatewayStreamError(message);
+  if (nested !== undefined) {
+    return errorFromOpenAIResponsesEvent(
+      'OpenAI Responses malformed stream error',
+      nested.code,
+      nested.message,
+      nested.param,
+    );
+  }
+
+  return errorFromOpenAIResponsesEvent(
+    'OpenAI Responses malformed stream error',
+    null,
+    message,
+    null,
+  );
+}
+
 function readResponsesFailedResponseError(response: RawObject):
   | {
       code: string | null;
@@ -699,7 +764,16 @@ export class OpenAIResponsesStreamedMessage implements StreamedMessage {
 
     try {
       for await (const chunk of response) {
-        const type = requireStringField(chunk, 'type', 'stream event');
+        const type = readStringField(chunk, 'type');
+        if (type === undefined) {
+          if (!hasOwn(chunk, 'type')) {
+            const message = readStringField(chunk, 'message');
+            if (message !== undefined) {
+              throw malformedStreamErrorEvent(message);
+            }
+          }
+          failResponsesDecode('stream event.type', 'must be a string.');
+        }
 
         switch (type) {
           case 'response.output_text.delta':
diff --git a/packages/kosong/test/openai-responses.test.ts b/packages/kosong/test/openai-responses.test.ts
index 572527cba4..3cfbd41ca8 100644
--- a/packages/kosong/test/openai-responses.test.ts
+++ b/packages/kosong/test/openai-responses.test.ts
@@ -1,4 +1,4 @@
-import { APIContextOverflowError, ChatProviderError } from '#/errors';
+import { APIContextOverflowError, APIStatusError, ChatProviderError } from '#/errors';
 import { generate } from '#/generate';
 import type { ContentPart, Message, StreamedMessagePart, ToolCall } from '#/message';
 import {
@@ -1682,8 +1682,52 @@ describe('OpenAIResponsesChatProvider', () => {
       ];
       const stream = new OpenAIResponsesStreamedMessage(makeAsyncIterable(events), true);
 
+      let caughtError: unknown;
+      try {
+        await collectStreamParts(stream);
+      } catch (error) {
+        caughtError = error;
+      }
+
+      expect(caughtError).toBeInstanceOf(APIStatusError);
+      expect((caughtError as APIStatusError).statusCode).toBe(429);
+      expect((caughtError as Error).message).toMatch(/rate_limit_exceeded.*too many/);
+    });
+
+    it('normalizes malformed gateway error frames with nested rate-limit JSON', async () => {
+      const events = [
+        {
+          message:
+            'received error while streaming: {"type":"tokens","code":"rate_limit_exceeded","message":"Rate limit reached for gpt-5.5. Please try again in 325ms.","param":null}',
+        },
+      ];
+      const stream = new OpenAIResponsesStreamedMessage(makeAsyncIterable(events), true);
+
+      let caughtError: unknown;
+      try {
+        await collectStreamParts(stream);
+      } catch (error) {
+        caughtError = error;
+      }
+
+      expect(caughtError).toBeInstanceOf(APIStatusError);
+      expect((caughtError as APIStatusError).statusCode).toBe(429);
+      expect((caughtError as Error).message).toContain('Rate limit reached for gpt-5.5');
+      expect((caughtError as Error).message).not.toContain('stream event.type must be a string');
+    });
+
+    it('rejects malformed stream events with a non-string type even when message is present', async () => {
+      const events = [
+        {
+          type: 42,
+          message:
+            'received error while streaming: {"type":"tokens","code":"rate_limit_exceeded","message":"too many requests","param":null}',
+        },
+      ];
+      const stream = new OpenAIResponsesStreamedMessage(makeAsyncIterable(events), true);
+
       await expect(collectStreamParts(stream)).rejects.toThrow(
-        /rate_limit_exceeded.*too many/,
+        'OpenAI Responses decode error: stream event.type must be a string.',
       );
     });