fix(responses): drop legacy cache retention for GPT-5.6

src/adapters/openai-responses.ts

@@ -185,6 +185,20 @@ function stripUnsupportedReasoningParams(body: unknown): unknown {
   return { ...body, reasoning: Object.keys(rest).length > 0 ? rest : undefined };
 }
 
+/**
+ * GPT-5.6 replaced the legacy 24-hour retention field with `prompt_cache_options.ttl`.
+ * Do not translate `24h` to the new field: GPT-5.6 currently accepts a different TTL contract,
+ * and implicit caching remains available when the caller did not send the replacement options.
+ */
+function stripDeprecatedPromptCacheRetention(body: unknown, modelId: unknown): unknown {
+  if (!isPlainObject(body)) return body;
+  if (typeof modelId !== "string") return body;
+  if (modelId !== "gpt-5.6" && !modelId.startsWith("gpt-5.6-")) return body;
+  if (!Object.hasOwn(body, "prompt_cache_retention")) return body;
+  const { prompt_cache_retention: _retention, ...rest } = body;
+  return rest;
+}
+
 /**
  * A false model capability prevents Codex from emitting summary fields after the catalog refresh.
  * Strip them here as well so an already-running client with a stale catalog cannot keep sending an
@@ -1491,6 +1505,9 @@ export function createResponsesPassthroughAdapter(provider: OcxProviderConfig):
       }
       if (forward) {
         outBody = stripUnsupportedForwardParams(outBody);
+        if (isCanonicalOpenAiForwardProvider(provider)) {
+          outBody = stripDeprecatedPromptCacheRetention(outBody, parsed.modelId);
+        }
       } else {
         outBody = preferConfiguredHostedTools(
           outBody,

tests/openai-responses-passthrough.test.ts

@@ -822,6 +822,57 @@ describe("OpenAI Responses passthrough sanitization", () => {
     expect(body.prompt_cache_retention).toBe("24h");
   });
 
+  test.each([
+    "gpt-5.6",
+    "gpt-5.6-sol",
+    "gpt-5.6-terra",
+    "gpt-5.6-luna",
+  ])("drops deprecated prompt_cache_retention for %s without inventing replacement options", modelId => {
+    const adapter = createResponsesPassthroughAdapter(provider);
+    const request = adapter.buildRequest({
+      modelId,
+      context: { messages: [] },
+      stream: true,
+      options: {},
+      _rawBody: {
+        model: modelId,
+        input: "hi",
+        prompt_cache_retention: "24h",
+      },
+    }, { headers: new Headers({ authorization: "Bearer token" }) });
+    const body = JSON.parse(request.body) as {
+      prompt_cache_retention?: string;
+      prompt_cache_options?: unknown;
+    };
+
+    expect(body.prompt_cache_retention).toBeUndefined();
+    expect(body.prompt_cache_options).toBeUndefined();
+  });
+
+  test("preserves caller prompt_cache_options while dropping GPT-5.6 legacy retention", () => {
+    const adapter = createResponsesPassthroughAdapter(provider);
+    const promptCacheOptions = { mode: "explicit", ttl: "30m" };
+    const request = adapter.buildRequest({
+      modelId: "gpt-5.6-sol",
+      context: { messages: [] },
+      stream: true,
+      options: {},
+      _rawBody: {
+        model: "gpt-5.6-sol",
+        input: "hi",
+        prompt_cache_retention: "24h",
+        prompt_cache_options: promptCacheOptions,
+      },
+    }, { headers: new Headers({ authorization: "Bearer token" }) });
+    const body = JSON.parse(request.body) as {
+      prompt_cache_retention?: string;
+      prompt_cache_options?: unknown;
+    };
+
+    expect(body.prompt_cache_retention).toBeUndefined();
+    expect(body.prompt_cache_options).toEqual(promptCacheOptions);
+  });
+
   const expandedRawBody = {
     model: "gpt-5.5",
     previous_response_id: "resp_1",
@@ -2017,6 +2068,52 @@ describe("OpenAI Responses forward-mode unsupported param stripping", () => {
     expect(body.max_output_tokens).toBe(32000);
     expect(body.metadata).toEqual({ user_id: "u-1" });
   });
+
+  test("key-auth custom endpoints preserve GPT-5.6 prompt_cache_retention", () => {
+    const adapter = createResponsesPassthroughAdapter({
+      adapter: "openai-responses",
+      baseUrl: "https://api.openai.example/v1",
+      authMode: "key",
+      apiKey: "sk-test",
+    });
+    const request = adapter.buildRequest({
+      modelId: "gpt-5.6-sol",
+      context: { messages: [] },
+      stream: true,
+      options: {},
+      _rawBody: {
+        model: "gpt-5.6-sol",
+        input: "hi",
+        prompt_cache_retention: "24h",
+      },
+    }, { headers: new Headers() });
+    const body = JSON.parse(request.body) as { prompt_cache_retention?: string };
+
+    expect(body.prompt_cache_retention).toBe("24h");
+  });
+
+  test("noncanonical forward endpoints preserve GPT-5.6 prompt_cache_retention", () => {
+    const adapter = createResponsesPassthroughAdapter({
+      adapter: "openai-responses",
+      baseUrl: "https://provider.example/v1",
+      authMode: "forward",
+      headers: { authorization: "Bearer provider-static" },
+    });
+    const request = adapter.buildRequest({
+      modelId: "gpt-5.6-sol",
+      context: { messages: [] },
+      stream: true,
+      options: {},
+      _rawBody: {
+        model: "gpt-5.6-sol",
+        input: "hi",
+        prompt_cache_retention: "24h",
+      },
+    }, { headers: new Headers() });
+    const body = JSON.parse(request.body) as { prompt_cache_retention?: string };
+
+    expect(body.prompt_cache_retention).toBe("24h");
+  });
 });
 
 describe("openaiResponsesUrl", () => {