diff --git a/packages/ai/src/providers/openai-completions.ts b/packages/ai/src/providers/openai-completions.ts
index 99e2e9134..f200c268c 100644
--- a/packages/ai/src/providers/openai-completions.ts
+++ b/packages/ai/src/providers/openai-completions.ts
@@ -156,16 +156,16 @@ function firstPositiveNumber(...values: unknown[]): number {
 	return 0;
 }
 
-function hasCacheReadTokenField(rawUsage: object): boolean {
+function hasPositiveCacheReadTokenField(rawUsage: object): boolean {
 	const usageLike = rawUsage as OpenAICompletionsUsageLike;
-	if (typeof usageLike.cached_tokens === "number") return true;
-	if (typeof usageLike.prompt_cache_hit_tokens === "number") return true;
+	if (typeof usageLike.cached_tokens === "number" && usageLike.cached_tokens > 0) return true;
+	if (typeof usageLike.prompt_cache_hit_tokens === "number" && usageLike.prompt_cache_hit_tokens > 0) return true;
 
 	const rawPromptTokenDetails = usageLike.prompt_tokens_details;
 	if (typeof rawPromptTokenDetails !== "object" || rawPromptTokenDetails === null) return false;
 
 	const promptTokenDetails = rawPromptTokenDetails as OpenAICompletionsPromptTokenDetails;
-	return typeof promptTokenDetails.cached_tokens === "number";
+	return typeof promptTokenDetails.cached_tokens === "number" && promptTokenDetails.cached_tokens > 0;
 }
 
 /**
@@ -1012,7 +1012,7 @@ const streamOpenAICompletionsOnce = (
 			const applyUsagePayload = (rawUsage: object): void => {
 				output.usage = parseChunkUsage(rawUsage, model, premiumRequestsTotal);
 				sawUsagePayload = true;
-				awaitTrailingUsageDetails = !hasCacheReadTokenField(rawUsage);
+				awaitTrailingUsageDetails = !hasPositiveCacheReadTokenField(rawUsage);
 			};
 			const timedOpenaiStream = iterateWithIdleTimeout(openaiStream, {
 				idleTimeoutMs,
diff --git a/packages/ai/test/openai-stream-terminal-close.test.ts b/packages/ai/test/openai-stream-terminal-close.test.ts
index 6f867d3a0..254ef4d2c 100644
--- a/packages/ai/test/openai-stream-terminal-close.test.ts
+++ b/packages/ai/test/openai-stream-terminal-close.test.ts
@@ -88,11 +88,16 @@ describe("terminal frame without connection close", () => {
 		expect(Date.now() - startedAt).toBeLessThan(2_000);
 	}, 10_000);
 
-	it("openai-completions: keeps reading usage-only cache details after finish usage", async () => {
+	it("openai-completions: ignores zero cache placeholder until trailing positive cache details arrive", async () => {
 		const fetchMock = createNeverClosingFetch([
 			completionChunk({
 				choices: [{ index: 0, delta: { role: "assistant", content: "Hello" }, finish_reason: "stop" }],
-				usage: { prompt_tokens: 10, completion_tokens: 5, total_tokens: 15 },
+				usage: {
+					prompt_tokens: 10,
+					completion_tokens: 5,
+					total_tokens: 15,
+					prompt_tokens_details: { cached_tokens: 0 },
+				},
 			}),
 			completionChunk({
 				choices: [],