diff --git a/extensions/openai/video-generation-provider.test.ts b/extensions/openai/video-generation-provider.test.ts
index f73e1f87bf79..02575dc3013b 100644
--- a/extensions/openai/video-generation-provider.test.ts
+++ b/extensions/openai/video-generation-provider.test.ts
@@ -11,6 +11,7 @@ const {
   fetchWithTimeoutMock,
   fetchWithTimeoutGuardedMock,
   pollProviderOperationJsonMock,
+  assertOkOrThrowHttpErrorMock,
   resolveProviderHttpRequestConfigMock,
   sanitizeConfiguredModelProviderRequestMock,
 } = getProviderHttpMocks();
@@ -303,6 +304,60 @@ describe("openai video generation provider", () => {
     });
   });
 
+  it("releases guarded local video download requests when HTTP errors throw", async () => {
+    const release = vi.fn(async () => {});
+    assertOkOrThrowHttpErrorMock
+      .mockImplementationOnce(async () => {})
+      .mockImplementationOnce(async () => {})
+      .mockImplementationOnce(async (_response, label) => {
+        throw new Error(label);
+      });
+    postJsonRequestMock.mockResolvedValue({
+      response: {
+        json: async () => ({
+          id: "vid_local",
+          model: "sora-2",
+          status: "queued",
+        }),
+      },
+      release: vi.fn(async () => {}),
+    });
+    fetchWithTimeoutMock.mockResolvedValueOnce({
+      json: async () => ({
+        id: "vid_local",
+        model: "sora-2",
+        status: "completed",
+      }),
+    });
+    fetchWithTimeoutGuardedMock.mockResolvedValueOnce({
+      response: new Response("busy", { status: 503, statusText: "Service Unavailable" }),
+      finalUrl: "http://127.0.0.1:44080/v1/videos/vid_local/content?variant=video",
+      release,
+    });
+
+    const provider = buildOpenAIVideoGenerationProvider();
+    await expect(
+      provider.generateVideo({
+        provider: "openai",
+        model: "sora-2",
+        prompt: "Render via local relay",
+        cfg: {
+          models: {
+            providers: {
+              openai: {
+                baseUrl: "http://127.0.0.1:44080/v1",
+                request: { allowPrivateNetwork: true },
+                models: [],
+              },
+            },
+          },
+        },
+      }),
+    ).rejects.toThrow("OpenAI video download failed");
+
+    expect(release).toHaveBeenCalledTimes(1);
+  });
+
   it("uses multipart input_reference for video-to-video uploads", async () => {
     fetchWithTimeoutMock
       .mockResolvedValueOnce({
diff --git a/extensions/openai/video-generation-provider.ts b/extensions/openai/video-generation-provider.ts
index ff5dc989cb00..c7d2a7e4772b 100644
--- a/extensions/openai/video-generation-provider.ts
+++ b/extensions/openai/video-generation-provider.ts
@@ -198,8 +198,13 @@ async function fetchOpenAIVideoDownload(
       auditContext: "openai-video-download",
     },
   );
-  await assertOkOrThrowHttpError(result.response, "OpenAI video download failed");
-  return result;
+  try {
+    await assertOkOrThrowHttpError(result.response, "OpenAI video download failed");
+    return result;
+  } catch (error) {
+    await result.release();
+    throw error;
+  }
 }
 
 async function downloadOpenAIVideo(
diff --git a/src/media-understanding/shared.test.ts b/src/media-understanding/shared.test.ts
index 9ce32876cbe7..c393a75b5b09 100644
--- a/src/media-understanding/shared.test.ts
+++ b/src/media-understanding/shared.test.ts
@@ -188,6 +188,48 @@ describe("provider operation deadlines", () => {
     expect(release).toHaveBeenCalledTimes(1);
   });
 
+  it("retries guarded transient provider status failures while polling", async () => {
+    vi.useFakeTimers();
+    vi.setSystemTime(1_000);
+    const firstRelease = vi.fn(async () => {});
+    const secondRelease = vi.fn(async () => {});
+    fetchWithSsrFGuardMock
+      .mockResolvedValueOnce({
+        response: new Response("busy", { status: 503, statusText: "Service Unavailable" }),
+        finalUrl: "https://api.example.com/v1/videos/task-1",
+        release: firstRelease,
+      })
+      .mockResolvedValueOnce({
+        response: new Response(JSON.stringify({ status: "completed" })),
+        finalUrl: "https://api.example.com/v1/videos/task-1",
+        release: secondRelease,
+      });
+
+    const result = pollProviderOperationJson<{ status?: string }>({
+      url: "https://api.example.com/v1/videos/task-1",
+      headers: new Headers({ authorization: "Bearer test" }),
+      deadline: createProviderOperationDeadline({
+        label: "video generation task task-1",
+        timeoutMs: 10_000,
+      }),
+      defaultTimeoutMs: 5_000,
+      fetchFn: fetch,
+      maxAttempts: 3,
+      pollIntervalMs: 1_000,
+      requestFailedMessage: "status failed",
+      timeoutMessage: "task timed out",
+      allowPrivateNetwork: true,
+      isComplete: (payload) => payload.status === "completed",
+    });
+
+    await vi.advanceTimersByTimeAsync(250);
+
+    await expect(result).resolves.toEqual({ status: "completed" });
+    expect(fetchWithSsrFGuardMock).toHaveBeenCalledTimes(2);
+    expect(firstRelease).toHaveBeenCalledTimes(1);
+    expect(secondRelease).toHaveBeenCalledTimes(1);
+  });
+
   it("throws provider failure messages while polling status JSON", async () => {
     const fetchFn = vi
       .fn<typeof fetch>()
diff --git a/src/media-understanding/shared.ts b/src/media-understanding/shared.ts
index a24ea72f9901..a77f4801c315 100644
--- a/src/media-understanding/shared.ts
+++ b/src/media-understanding/shared.ts
@@ -174,15 +174,16 @@ export async function pollProviderOperationJson<TPayload>(
     const guardedOptions = resolveGuardedRequestOptions(params);
     const payload = guardedOptions
       ? await (async () => {
-          const result = await fetchWithTimeoutGuarded(
-            params.url,
+          const result = await fetchGuardedProviderOperationResponse({
+            stage: "poll",
+            url: params.url,
             init,
-            timeoutMs(),
-            params.fetchFn,
+            timeoutMs,
+            fetchFn: params.fetchFn,
+            requestFailedMessage: params.requestFailedMessage,
             guardedOptions,
-          );
+          });
           try {
-            await assertOkOrThrowHttpError(result.response, params.requestFailedMessage);
             return (await readProviderJsonObjectResponse(
               result.response,
               params.requestFailedMessage,
@@ -473,6 +474,42 @@ function resolveGuardedRequestOptions(
   };
 }
 
+async function fetchGuardedProviderOperationResponse(params: {
+  stage: ProviderOperationRetryStage;
+  url: string;
+  init: RequestInit;
+  timeoutMs?: ProviderOperationTimeoutMs;
+  fetchFn: typeof fetch;
+  provider?: string;
+  requestFailedMessage?: string;
+  retry?: TransientProviderRetryConfig;
+  guardedOptions: GuardedProviderRequestOptions;
+}): Promise<GuardedFetchResult> {
+  return await executeProviderOperationWithRetry({
+    provider: params.provider ?? "provider-http",
+    stage: params.stage,
+    retry: params.retry,
+    operation: async () => {
+      const result = await fetchWithTimeoutGuarded(
+        params.url,
+        params.init,
+        resolveProviderOperationRequestTimeoutMs(params.timeoutMs),
+        params.fetchFn,
+        params.guardedOptions,
+      );
+      try {
+        if (params.requestFailedMessage) {
+          await assertOkOrThrowHttpError(result.response, params.requestFailedMessage);
+        }
+        return result;
+      } catch (error) {
+        await result.release();
+        throw error;
+      }
+    },
+  });
+}
+
 type GuardedPostRequestRetryOptions = {
   /**
    * POST requests default to no retry because many provider endpoints create