diff --git a/src/providers/request-pacing.ts b/src/providers/request-pacing.ts
index e38c62f3c0..5b5f34d3f6 100644
--- a/src/providers/request-pacing.ts
+++ b/src/providers/request-pacing.ts
@@ -35,7 +35,7 @@ interface Waiter {
   modelIntervalMs: number;
   queuedAt: number;
   signal?: AbortSignal;
-  resolve: () => void;
+  start: () => void;
   reject: (reason: unknown) => void;
   abort?: () => void;
 }
@@ -190,19 +190,29 @@ function runQueue(providerName: string, state: ProviderPacer): void {
   if (waiter.modelId && waiter.modelIntervalMs > 0) {
     state.modelNextStartAt.set(waiter.modelId, startedAt + waiter.modelIntervalMs);
   }
-  waiter.resolve();
+  // Start the caller-owned transport synchronously in the granted slot. Resolving a
+  // waiter first would let an async continuation drift later while the next slot
+  // remained anchored to this earlier bookkeeping timestamp.
+  waiter.start();
   runtime.enqueueMicrotask(() => runQueue(providerName, state));
 }
 
-export async function waitForProviderRequestSlot(
+export function runWithProviderRequestSlot<T>(
   providerName: string,
   provider: OcxProviderConfig,
-  modelId?: string,
-  signal?: AbortSignal,
-): Promise<void> {
+  modelId: string | undefined,
+  signal: AbortSignal | undefined,
+  start: () => T | PromiseLike<T>,
+): Promise<T> {
   const intervals = requestPacingIntervals(provider, modelId);
-  if (Math.max(intervals.providerIntervalMs, intervals.modelIntervalMs) <= 0) return;
-  if (signal?.aborted) throw abortReason(signal);
+  if (signal?.aborted) return Promise.reject(abortReason(signal));
+  if (Math.max(intervals.providerIntervalMs, intervals.modelIntervalMs) <= 0) {
+    try {
+      return Promise.resolve(start());
+    } catch (error) {
+      return Promise.reject(error);
+    }
+  }
 
   const state = pacers.get(providerName) ?? {
     queue: [], providerNextStartAt: 0, modelNextStartAt: new Map<string, number>(),
@@ -218,15 +228,28 @@ export async function waitForProviderRequestSlot(
   }
   runQueue(providerName, state);
   if (state.queue.length >= maxQueueDepth) {
-    throw new RequestPacingQueueOverloadError(
+    return Promise.reject(new RequestPacingQueueOverloadError(
       providerName,
       "queue_full",
       pacingRetryAfterSeconds(state, modelId, runtime.now()),
-    );
+    ));
   }
 
-  await new Promise<void>((resolve, reject) => {
-    const waiter: Waiter = { modelId, ...intervals, queuedAt: runtime.now(), signal, resolve, reject };
+  return new Promise<T>((resolve, reject) => {
+    const waiter: Waiter = {
+      modelId,
+      ...intervals,
+      queuedAt: runtime.now(),
+      signal,
+      reject,
+      start: () => {
+        try {
+          resolve(start());
+        } catch (error) {
+          reject(error);
+        }
+      },
+    };
     waiter.abort = () => {
       const index = state.queue.indexOf(waiter);
       if (index >= 0) state.queue.splice(index, 1);
@@ -252,6 +275,15 @@ export async function waitForProviderRequestSlot(
   });
 }
 
+export function waitForProviderRequestSlot(
+  providerName: string,
+  provider: OcxProviderConfig,
+  modelId?: string,
+  signal?: AbortSignal,
+): Promise<void> {
+  return runWithProviderRequestSlot(providerName, provider, modelId, signal, () => undefined);
+}
+
 export function providerRequestPacingStatus(
   providerName: string,
   provider: OcxProviderConfig,
diff --git a/src/server/responses/fetch-helpers.ts b/src/server/responses/fetch-helpers.ts
index 21652a5345..01cf3badb9 100644
--- a/src/server/responses/fetch-helpers.ts
+++ b/src/server/responses/fetch-helpers.ts
@@ -102,7 +102,7 @@ import {
 } from "../relay";
 import { hasResponsesItemIdRepair, relaySseWithResponsesItemIdRepair } from "../responses-item-id-repair";
 import type { EffectiveSubagentRoster, SpawnAgentSurface } from "../../codex/catalog";
-import { waitForProviderRequestSlot } from "../../providers/request-pacing";
+import { runWithProviderRequestSlot, waitForProviderRequestSlot } from "../../providers/request-pacing";
 
 
 export function disableResponsesRequestTimeout(req: Request, server: Pick<Server<WsData>, "timeout"> | undefined): boolean {
@@ -139,6 +139,7 @@ export function safeOriginLabel(url: string): string {
 
 export interface PaceAwareFetch {
   waitForPacing?: (signal?: AbortSignal) => Promise<void>;
+  runWithPacing?: <T>(signal: AbortSignal | undefined, start: () => T | PromiseLike<T>) => Promise<T>;
   unpacedFetch?: typeof globalThis.fetch;
 }
 
@@ -199,16 +200,26 @@ export function providerFetch(
   const waitForPacing = (signal?: AbortSignal) => options.providerName
     ? waitForProviderRequestSlot(options.providerName, provider, options.modelId, signal)
     : Promise.resolve();
-  const wrapped = async (input: Parameters<typeof globalThis.fetch>[0], init?: RequestInit) => {
-    await waitForPacing(init?.signal ?? undefined);
-    return unpaced(input, init);
+  const runWithPacing = <T>(signal: AbortSignal | undefined, start: () => T | PromiseLike<T>): Promise<T> => {
+    if (options.providerName) {
+      return runWithProviderRequestSlot(options.providerName, provider, options.modelId, signal, start);
+    }
+    if (signal?.aborted) return Promise.reject(signal.reason ?? new DOMException("The operation was aborted", "AbortError"));
+    try {
+      return Promise.resolve(start());
+    } catch (error) {
+      return Promise.reject(error);
+    }
   };
+  const wrapped = (input: Parameters<typeof globalThis.fetch>[0], init?: RequestInit) =>
+    runWithPacing(init?.signal ?? undefined, () => unpaced(input, init));
   const preconnect = (...args: Parameters<typeof globalThis.fetch.preconnect>): void => {
     base.preconnect?.(...args);
   };
   return Object.assign(wrapped, {
     preconnect,
     waitForPacing,
+    runWithPacing,
     unpacedFetch: Object.assign(unpaced, { preconnect }),
   });
 }
@@ -225,29 +236,34 @@ export async function fetchWithHeaderTimeout(
   manualRedirect = false,
 ): Promise<Response> {
   const pacing = executor as ProviderFetch;
-  await pacing.waitForPacing?.(abortSignal);
   const fetchExecutor = pacing.unpacedFetch ?? executor;
-  const timeout = new AbortController();
-  const timer = setTimeout(() => {
-    if (!timeout.signal.aborted) timeout.abort(new DOMException("Timeout elapsed", "TimeoutError"));
-  }, timeoutMs);
-  const headers = new Headers(init.headers);
-  // Compressed SSE can be held until the decompressor has a complete block. Streaming calls
-  // default to identity for low-latency frame delivery, while an explicit caller choice wins.
-  if (preferIdentityEncoding && !headers.has("accept-encoding")) {
-    headers.set("accept-encoding", "identity");
-  }
-  try {
-    return await fetchExecutor(url, {
-      ...init,
-      headers,
-      // Credential-bearing sends opt into manual redirects so a 3xx is relayed
-      // as a Response instead of being followed into a rejection that is
-      // indistinguishable from a pre-connection failure (#914).
-      ...(manualRedirect ? { redirect: "manual" as const } : {}),
-      signal: AbortSignal.any([abortSignal, timeout.signal]),
-    });
-  } finally {
-    clearTimeout(timer);
-  }
+  const startFetch = async (): Promise<Response> => {
+    const timeout = new AbortController();
+    const timer = setTimeout(() => {
+      if (!timeout.signal.aborted) timeout.abort(new DOMException("Timeout elapsed", "TimeoutError"));
+    }, timeoutMs);
+    const headers = new Headers(init.headers);
+    // Compressed SSE can be held until the decompressor has a complete block. Streaming calls
+    // default to identity for low-latency frame delivery, while an explicit caller choice wins.
+    if (preferIdentityEncoding && !headers.has("accept-encoding")) {
+      headers.set("accept-encoding", "identity");
+    }
+    try {
+      return await fetchExecutor(url, {
+        ...init,
+        headers,
+        // Credential-bearing sends opt into manual redirects so a 3xx is relayed
+        // as a Response instead of being followed into a rejection that is
+        // indistinguishable from a pre-connection failure (#914).
+        ...(manualRedirect ? { redirect: "manual" as const } : {}),
+        signal: AbortSignal.any([abortSignal, timeout.signal]),
+      });
+    } finally {
+      clearTimeout(timer);
+    }
+  };
+
+  if (pacing.runWithPacing) return pacing.runWithPacing(abortSignal, startFetch);
+  await pacing.waitForPacing?.(abortSignal);
+  return startFetch();
 }
diff --git a/tests/request-pacing.test.ts b/tests/request-pacing.test.ts
index 4a2dedc60a..23157270aa 100644
--- a/tests/request-pacing.test.ts
+++ b/tests/request-pacing.test.ts
@@ -84,10 +84,12 @@ describe("requestPacingIntervalMs", () => {
 });
 
 describe("provider request pacing queue", () => {
-  test("spaces concurrent starts in one provider FIFO and exposes queue state", async () => {
+  test("anchors concurrent pacing to actual transport starts and exposes queue state", async () => {
+    const clock = fakePacingClock();
+    setProviderRequestPacingRuntimeForTest(clock.runtime);
     const starts: number[] = [];
     const fetchImpl = Object.assign(async () => {
-      starts.push(Date.now());
+      starts.push(clock.now());
       return new Response("ok");
     }, { preconnect() {} }) as typeof globalThis.fetch;
     const configured = {
@@ -96,12 +98,17 @@ describe("provider request pacing queue", () => {
     } as OcxProviderConfig & { fetch: typeof globalThis.fetch };
     const send = providerFetch(configured, undefined, { providerName: "demo", modelId: "model-a" });
     const pending = [send("https://example.test/v1/chat/completions"), send("https://example.test/v1/chat/completions"), send("https://example.test/v1/chat/completions")];
-    await Bun.sleep(10);
+
+    expect(starts).toEqual([0]);
     expect(providerRequestPacingStatus("demo", configured).queued).toBe(2);
+    clock.advanceBy(99);
+    expect(starts).toEqual([0]);
+    clock.advanceBy(1);
+    expect(starts).toEqual([0, 100]);
+    clock.advanceBy(100);
+    expect(starts).toEqual([0, 100, 200]);
+
     await Promise.all(pending);
-    expect(starts).toHaveLength(3);
-    expect(starts[1] - starts[0]).toBeGreaterThanOrEqual(85);
-    expect(starts[2] - starts[1]).toBeGreaterThanOrEqual(85);
     const status = providerRequestPacingStatus("demo", configured);
     expect(status.queued).toBe(0);
     expect(status.lastModelId).toBe("model-a");