diff --git a/CHANGELOG.md b/CHANGELOG.md
index a489109..3be492a 100644
--- a/CHANGELOG.md
+++ b/CHANGELOG.md
@@ -6,6 +6,13 @@ See [ARCHITECTURE.md](ARCHITECTURE.md) for current direction and
 
 ## Unreleased
 
+- **Stop cutting Claude responses at three minutes.** The gateway no longer applies
+  an implicit 180-second deadline to Claude passthrough, matching the OpenAI and Zen
+  routes. Long streamed Claude turns, including server-side advisor calls, previously
+  ended mid-response ("Server error mid-response"; `claude-multi -p` reported the
+  truncated text as success). Client disconnects and explicit gateway timeouts still
+  abort upstream requests.
+
 ## 0.2.1 — 2026-09-22
 
 - Explain missing native worker spawn acknowledgements with the unsupported
diff --git a/plugins/multi-core/src/gateway/server.ts b/plugins/multi-core/src/gateway/server.ts
index ac2ea5f..a202ecd 100644
--- a/plugins/multi-core/src/gateway/server.ts
+++ b/plugins/multi-core/src/gateway/server.ts
@@ -887,13 +887,12 @@ function providerSignal(disconnected: AbortSignal, model: string | null, timeout
   ) {
     return disconnected;
   }
-  if (
-    (model?.startsWith('multi/openai/') || model?.startsWith('multi/zen/')) &&
-    timeoutMs === undefined
-  ) {
+  // Streamed OpenAI, Zen and Claude responses (including server-side advisor calls)
+  // can outlive any fixed wall-clock deadline; only an explicit limit bounds them.
+  if (timeoutMs === undefined) {
     return disconnected;
   }
-  return AbortSignal.any([disconnected, AbortSignal.timeout(timeoutMs ?? 180000)]);
+  return AbortSignal.any([disconnected, AbortSignal.timeout(timeoutMs)]);
 }
 
 async function readRequest(req: http.IncomingMessage, catalog?: AgentCatalog) {
diff --git a/test/unit/native-gateway.test.ts b/test/unit/native-gateway.test.ts
index 211a94d..8510754 100644
--- a/test/unit/native-gateway.test.ts
+++ b/test/unit/native-gateway.test.ts
@@ -1246,7 +1246,7 @@ test('oversized uploads receive HTTP 413 while the client is still streaming', a
   assert.equal(await response.text(), 'Request too large');
 });
 
-test('OpenAI has no implicit request deadline while explicit limits and Claude passthrough remain bounded', async (t) => {
+test('OpenAI and Claude passthrough have no implicit request deadline while explicit limits bound both', async (t) => {
   const durations: number[] = [];
   t.mock.method(AbortSignal, 'timeout', (ms: number) => {
     durations.push(ms);
@@ -1260,10 +1260,15 @@ test('OpenAI has no implicit request deadline while explicit limits and Claude p
   assert.equal((await ordinary(body)).status, 200);
   assert.deepEqual(durations, [], 'Astra must not inherit an absolute three-minute timer');
   assert.equal((await ordinary({ model: 'claude-sonnet-5', messages })).status, 200);
-  assert.deepEqual(durations, [180000]);
+  assert.deepEqual(
+    durations,
+    [],
+    'Claude passthrough must not cut long streams, such as advisor calls, at three minutes',
+  );
   const bounded = await gateway(t, upstream, { timeoutMs: 25 });
   assert.equal((await bounded(body)).status, 200);
-  assert.deepEqual(durations, [180000, 25]);
+  assert.equal((await bounded({ model: 'claude-sonnet-5', messages })).status, 200);
+  assert.deepEqual(durations, [25, 25]);
 });
 
 test('an explicit OpenAI timeout aborts upstream inference', async (t) => {