diff --git a/src/lib/responses/ResponseStream.ts b/src/lib/responses/ResponseStream.ts
index 1761f4bbf..563326787 100644
--- a/src/lib/responses/ResponseStream.ts
+++ b/src/lib/responses/ResponseStream.ts
@@ -27,7 +27,9 @@ export type ResponseStreamByIdParams = {
    */
   response_id: string;
   /**
-   * If provided, the stream will start after the event with the given sequence number.
+   * If provided, events with a sequence number less than or equal to this value
+   * will not be emitted. The helper still replays them internally to build a
+   * complete snapshot for later events and `finalResponse()`.
    */
   starting_after?: number;
   /**
@@ -175,6 +177,8 @@ export class ResponseStream<ParsedT = null>
     let stream: Stream<ResponseStreamEvent> | undefined;
     let starting_after: number | null = null;
     if ('response_id' in params) {
+      // Keep the full replay so that `accumulateResponse()` sees `response.created` and can build
+      // complete snapshots before locally filtering events at `starting_after`.
       stream = await client.responses.retrieve(
         params.response_id,
         { stream: true },
diff --git a/tests/lib/ResponseStream.test.ts b/tests/lib/ResponseStream.test.ts
index d23449b34..a656a3a30 100644
--- a/tests/lib/ResponseStream.test.ts
+++ b/tests/lib/ResponseStream.test.ts
@@ -1,8 +1,77 @@
 import { makeStreamSnapshotRequest } from '../utils/mock-snapshots';
+import OpenAI from 'openai';
 
 jest.setTimeout(1000 * 30);
 
 describe('.stream()', () => {
+  it('replays prior events when resuming by ID so snapshots stay complete', async () => {
+    const requests: string[] = [];
+    const response = {
+      id: 'resp_123',
+      object: 'response',
+      created_at: 0,
+      model: 'gpt-4o',
+      output: [],
+      error: null,
+      incomplete_details: null,
+      instructions: null,
+      metadata: null,
+      parallel_tool_calls: false,
+      temperature: null,
+      tools: [],
+      top_p: null,
+      status: 'completed',
+      usage: null,
+    };
+    const events = [
+      {
+        type: 'response.created',
+        sequence_number: 0,
+        response: { ...response, status: 'in_progress' },
+      },
+      {
+        type: 'response.output_item.added',
+        sequence_number: 1,
+        output_index: 0,
+        item: {
+          id: 'msg_1',
+          type: 'message',
+          role: 'assistant',
+          status: 'in_progress',
+          content: [],
+        },
+      },
+      { type: 'response.completed', sequence_number: 2, response },
+    ];
+    const openai = new OpenAI({
+      apiKey: 'My API Key',
+      fetch: async (url) => {
+        const requestURL = String(url);
+        requests.push(requestURL);
+        // Match the API's cursor behavior: forwarding `starting_after` omits the prefix needed
+        // by the accumulator, while omitting it replays the complete event sequence.
+        const streamEvents = requestURL.includes('starting_after=') ? events.slice(1) : events;
+        const body = `${streamEvents
+          .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' },
+        });
+      },
+    });
+
+    const emittedEvents: number[] = [];
+    const stream = openai.responses
+      .stream({ response_id: 'resp_123', starting_after: 0 })
+      .on('event', (event) => emittedEvents.push(event.sequence_number));
+    const final = await stream.finalResponse();
+
+    expect(requests).toEqual(['https://api.openai.com/v1/responses/resp_123?stream=true']);
+    expect(emittedEvents).toEqual([1, 2]);
+    expect(final.id).toBe('resp_123');
+  });
+
   it('standard text works', async () => {
     const deltas: string[] = [];