diff --git a/packages/core/src/core/openaiContentGenerator/pipeline.test.ts b/packages/core/src/core/openaiContentGenerator/pipeline.test.ts
index eaec7a7bdb..00a3d8c6bf 100644
--- a/packages/core/src/core/openaiContentGenerator/pipeline.test.ts
+++ b/packages/core/src/core/openaiContentGenerator/pipeline.test.ts
@@ -6154,8 +6154,14 @@ describe('ContentGenerationPipeline', () => {
       expect(err).toBeInstanceOf(StreamInactivityTimeoutError);
       expect((err as Error).message).toBe(
         'No stream activity for 1000ms after 0 chunks ' +
-          '(stream lifetime: 1000ms). Set QWEN_STREAM_IDLE_TIMEOUT_MS ' +
-          'to increase this window (or 0 to disable it).',
+          '(stream lifetime: 1000ms). For provider-backed models, ' +
+          'increase modelProviders[providerId][].generationConfig.streamIdleTimeoutMs; ' +
+          'provider configuration takes precedence, so model.generationConfig is ' +
+          'ignored for those models. For runtime models, increase ' +
+          'model.generationConfig.streamIdleTimeoutMs. Built-in Qwen OAuth models ' +
+          'cannot be overridden via settings. Use QWEN_STREAM_IDLE_TIMEOUT_MS ' +
+          'for them or whenever no explicit value is active. ' +
+          'Set the active value to 0 to disable it.',
       );
       expect(err).toMatchObject({ code: 'ETIMEDOUT' });
       expect((err as StreamInactivityTimeoutError).chunksReceived).toBe(0);
@@ -6164,7 +6170,7 @@ describe('ContentGenerationPipeline', () => {
       expect(mockErrorHandler.handle).not.toHaveBeenCalled();
     });
 
-    it('includes the idle detail and env override hint in timeout errors', async () => {
+    it('includes settings and environment override hints in timeout errors', async () => {
       const gated = gatedStream(); // never push/end → silent
       (mockClient.chat.completions.create as Mock).mockResolvedValue(
         gated.stream,
@@ -6184,6 +6190,10 @@ describe('ContentGenerationPipeline', () => {
       expect(err).toBeInstanceOf(StreamInactivityTimeoutError);
       const message = (err as Error).message;
       expect(message).toContain('No stream activity for 1000ms after 0 chunks');
+      expect(message).toContain('model.generationConfig.streamIdleTimeoutMs');
+      expect(message).toContain(
+        'modelProviders[providerId][].generationConfig.streamIdleTimeoutMs',
+      );
       expect(message).toContain('QWEN_STREAM_IDLE_TIMEOUT_MS');
     });
 
diff --git a/packages/core/src/core/openaiContentGenerator/pipeline.ts b/packages/core/src/core/openaiContentGenerator/pipeline.ts
index f3f8d89146..599d8234f6 100644
--- a/packages/core/src/core/openaiContentGenerator/pipeline.ts
+++ b/packages/core/src/core/openaiContentGenerator/pipeline.ts
@@ -191,9 +191,14 @@ export class StreamInactivityTimeoutError extends Error {
   ) {
     super(
       `No stream activity for ${idleMs}ms after ${chunksReceived} chunks ` +
-        `(stream lifetime: ${streamLifetimeMs}ms). Set ` +
-        `${QWEN_STREAM_IDLE_TIMEOUT_MS_ENV} to increase this window ` +
-        `(or 0 to disable it).`,
+        `(stream lifetime: ${streamLifetimeMs}ms). For provider-backed models, ` +
+        `increase modelProviders[providerId][].generationConfig.streamIdleTimeoutMs; ` +
+        `provider configuration takes precedence, so model.generationConfig is ` +
+        `ignored for those models. For runtime models, increase ` +
+        `model.generationConfig.streamIdleTimeoutMs. Built-in Qwen OAuth models ` +
+        `cannot be overridden via settings. Use ${QWEN_STREAM_IDLE_TIMEOUT_MS_ENV} ` +
+        `for them or whenever no explicit value is active. ` +
+        `Set the active value to 0 to disable it.`,
     );
     this.name = 'StreamInactivityTimeoutError';
   }
diff --git a/packages/core/src/models/content-generator-config.test.ts b/packages/core/src/models/content-generator-config.test.ts
index da4d2f3f17..5f3a81f5df 100644
--- a/packages/core/src/models/content-generator-config.test.ts
+++ b/packages/core/src/models/content-generator-config.test.ts
@@ -193,6 +193,24 @@ describe('buildAgentContentGeneratorConfig', () => {
       expect(result.extra_body).toBeUndefined();
     });
 
+    it('should preserve a zero stream idle timeout from the registry', () => {
+      const config = createMockConfig(parentConfig, {
+        ...resolvedModel,
+        generationConfig: {
+          ...resolvedModel.generationConfig,
+          streamIdleTimeoutMs: 0,
+        },
+      });
+
+      const result = buildAgentContentGeneratorConfig(
+        config,
+        'registry-model-id',
+        { authType: 'anthropic' },
+      );
+
+      expect(result.streamIdleTimeoutMs).toBe(0);
+    });
+
     it('should prefer explicit auth overrides over registry values', () => {
       const config = createMockConfig(parentConfig, resolvedModel);