diff --git a/open-sse/translator/request/openai-responses/toResponses.ts b/open-sse/translator/request/openai-responses/toResponses.ts
index d52608496f..bda9113d08 100644
--- a/open-sse/translator/request/openai-responses/toResponses.ts
+++ b/open-sse/translator/request/openai-responses/toResponses.ts
@@ -17,6 +17,16 @@ import {
   normalizeResponsesReasoningEffort,
 } from "./helpers.ts";
 
+// A Chat-Completions client can only express reasoning via the top-level
+// `reasoning_effort` hint; it has no way to request a reasoning summary. When we
+// promote that hint to the Responses API's `reasoning.effort`, default the
+// summary (plus the encrypted-content include that actually streams it) so a
+// downstream chat client still sees a thinking stream instead of an empty
+// summary. Mirrors the Codex executor's `ensureCodexReasoningSummary`. An
+// explicit `reasoning` object from a Responses-shaped client is preserved as-is.
+const DEFAULT_RESPONSES_REASONING_SUMMARY = "auto";
+const RESPONSES_REASONING_ENCRYPTED_CONTENT_INCLUDE = "reasoning.encrypted_content";
+
 // Chat Completions `response_format: { type: "json_schema" }` → Responses API `text.format`.
 // Merges into any existing `result.text` (e.g. verbosity) so structured-output schemas from
 // Chat clients survive the translation to the Responses/Codex upstream (#5933).
@@ -329,11 +339,17 @@ export function openaiToOpenAIResponsesRequest(
   if (chatVerbosity) {
     result.text = { ...toRecord(result.text), verbosity: chatVerbosity };
   }
+  let defaultedReasoningSummary = false;
   if (root.reasoning !== undefined) {
     result.reasoning = root.reasoning;
   } else if (root.reasoning_effort !== undefined) {
     const effort = normalizeResponsesReasoningEffort(root.reasoning_effort);
-    if (effort) {
+    if (effort && effort !== "none") {
+      // Effort-only chat request: default a reasoning summary so the upstream
+      // streams thinking back (see the constant's note above).
+      result.reasoning = { effort, summary: DEFAULT_RESPONSES_REASONING_SUMMARY };
+      defaultedReasoningSummary = true;
+    } else if (effort) {
       result.reasoning = { effort };
     }
   }
@@ -345,6 +361,14 @@ export function openaiToOpenAIResponsesRequest(
   if (Array.isArray(root.include) && root.include.length > 0) {
     result.include = root.include;
   }
+  // When we defaulted a reasoning summary above, also request the encrypted
+  // reasoning content so the summary actually streams back to the chat client.
+  if (defaultedReasoningSummary) {
+    const include = Array.isArray(result.include) ? (result.include as unknown[]) : [];
+    if (!include.includes(RESPONSES_REASONING_ENCRYPTED_CONTENT_INCLUDE)) {
+      result.include = [...include, RESPONSES_REASONING_ENCRYPTED_CONTENT_INCLUDE];
+    }
+  }
   if (storeEnabled) {
     if (root[RESPONSES_STORE_MARKER] !== undefined) {
       result.store = root[RESPONSES_STORE_MARKER];
diff --git a/tests/unit/openai-responses-reasoning-effort.test.ts b/tests/unit/openai-responses-reasoning-effort.test.ts
index 9b8151bf15..a8a2927d64 100644
--- a/tests/unit/openai-responses-reasoning-effort.test.ts
+++ b/tests/unit/openai-responses-reasoning-effort.test.ts
@@ -68,5 +68,5 @@ test("Chat -> Responses already wraps reasoning_effort into reasoning.effort", (
       {}
     )
   );
-  assert.deepEqual(out.reasoning, { effort: "high" });
+  assert.deepEqual(out.reasoning, { effort: "high", summary: "auto" });
 });
diff --git a/tests/unit/translator-openai-responses-req.test.ts b/tests/unit/translator-openai-responses-req.test.ts
index 87e5c97f43..3fdda8c31a 100644
--- a/tests/unit/translator-openai-responses-req.test.ts
+++ b/tests/unit/translator-openai-responses-req.test.ts
@@ -599,11 +599,34 @@ test("Chat -> Responses maps reasoning_effort into Responses reasoning", () => {
     null
   );
 
-  assert.deepEqual((result as any).reasoning, { effort: "low" });
+  // Effort-only chat requests now default `summary: "auto"` + the encrypted
+  // reasoning include so Responses-API upstreams stream thinking back to the
+  // chat client (previously the summary was empty and no think was visible).
+  assert.deepEqual((result as any).reasoning, { effort: "low", summary: "auto" });
+  assert.deepEqual((result as Record<string, unknown>).include, ["reasoning.encrypted_content"]);
   assert.equal((result as any).reasoning_effort, undefined);
   assert.equal((result as any).store, false);
 });
 
+test("Chat -> Responses does not default a reasoning summary for reasoning_effort none", () => {
+  const result = openaiToOpenAIResponsesRequest(
+    "gpt-5.3-codex-spark",
+    {
+      messages: [{ role: "user", content: "Hello" }],
+      reasoning_effort: "none",
+    },
+    false,
+    null
+  );
+
+  const record = result as Record<string, unknown>;
+  const reasoning = record.reasoning as Record<string, unknown> | undefined;
+  if (reasoning !== undefined) {
+    assert.equal(reasoning.summary, undefined);
+  }
+  assert.equal(record.include, undefined);
+});
+
 test("Chat -> Responses normalizes reasoning_effort max to xhigh", () => {
   const result = openaiToOpenAIResponsesRequest(
     "gpt-5.5",
@@ -615,7 +638,7 @@ test("Chat -> Responses normalizes reasoning_effort max to xhigh", () => {
     null
   );
 
-  assert.deepEqual((result as any).reasoning, { effort: "xhigh" });
+  assert.deepEqual((result as any).reasoning, { effort: "xhigh", summary: "auto" });
   assert.equal((result as any).reasoning_effort, undefined);
 });