fix: respect explicit model selection after Flash quota exhaustion (#26759) (#26872)

packages/core/src/routing/strategies/classifierStrategy.test.ts

@@ -27,6 +27,7 @@ import type { Content } from '@google/genai';
 import type { ResolvedModelConfig } from '../../services/modelConfigService.js';
 import { debugLogger } from '../../utils/debugLogger.js';
 import { AuthType } from '../../core/contentGenerator.js';
+import { ModelAvailabilityService } from '../../availability/modelAvailabilityService.js';
 
 vi.mock('../../core/baseLlmClient.js');
 
@@ -68,6 +69,9 @@ describe('ClassifierStrategy', () => {
       getContentGeneratorConfig: vi.fn().mockReturnValue({
         authType: AuthType.LOGIN_WITH_GOOGLE,
       }),
+      getModelAvailabilityService: vi
+        .fn()
+        .mockReturnValue(new ModelAvailabilityService()),
     } as unknown as Config;
     mockBaseLlmClient = {
       generateJson: vi.fn(),

packages/core/src/routing/strategies/classifierStrategy.ts

@@ -20,6 +20,7 @@ import {
   isFunctionResponse,
 } from '../../utils/messageInspectors.js';
 import { debugLogger } from '../../utils/debugLogger.js';
+import { normalizeModelId } from '../../utils/modelUtils.js';
 import type { LocalLiteRtLmClient } from '../../core/localLiteRtLmClient.js';
 import { LlmRole } from '../../telemetry/types.js';
 
@@ -177,16 +178,28 @@ export class ClassifierStrategy implements RoutingStrategy {
           config.getGemini31FlashLiteLaunched(),
           config.getUseCustomToolModel(),
         ]);
-      const selectedModel = resolveClassifierModel(
-        model,
-        routerResponse.model_choice,
-        useGemini3_1,
-        useGemini3_1FlashLite,
-        useCustomToolModel,
-        config.getHasAccessToPreviewModel?.() ?? true,
-        config,
+      const selectedModel = normalizeModelId(
+        resolveClassifierModel(
+          normalizeModelId(model),
+          routerResponse.model_choice,
+          useGemini3_1,
+          useGemini3_1FlashLite,
+          useCustomToolModel,
+          config.getHasAccessToPreviewModel?.() ?? true,
+          config,
+        ),
       );
 
+      const service = config.getModelAvailabilityService();
+      const snapshot = service.snapshot(selectedModel);
+
+      if (!snapshot.available) {
+        debugLogger.warn(
+          `[Routing] Classifier selected unavailable model ${selectedModel} (${snapshot.reason}). Bypassing.`,
+        );
+        return null;
+      }
+
       return {
         model: selectedModel,
         metadata: {

packages/core/src/routing/strategies/numericalClassifierStrategy.test.ts

@@ -27,6 +27,7 @@ import type { ResolvedModelConfig } from '../../services/modelConfigService.js';
 import { debugLogger } from '../../utils/debugLogger.js';
 import type { LocalLiteRtLmClient } from '../../core/localLiteRtLmClient.js';
 import { AuthType } from '../../core/contentGenerator.js';
+import { ModelAvailabilityService } from '../../availability/modelAvailabilityService.js';
 
 vi.mock('../../core/baseLlmClient.js');
 
@@ -71,6 +72,9 @@ describe('NumericalClassifierStrategy', () => {
       getContentGeneratorConfig: vi.fn().mockReturnValue({
         authType: AuthType.LOGIN_WITH_GOOGLE,
       }),
+      getModelAvailabilityService: vi
+        .fn()
+        .mockReturnValue(new ModelAvailabilityService()),
     } as unknown as Config;
     mockBaseLlmClient = {
       generateJson: vi.fn(),

packages/core/src/routing/strategies/numericalClassifierStrategy.ts

@@ -20,6 +20,7 @@ import {
   isFunctionResponse,
 } from '../../utils/messageInspectors.js';
 import { debugLogger } from '../../utils/debugLogger.js';
+import { normalizeModelId } from '../../utils/modelUtils.js';
 import type { LocalLiteRtLmClient } from '../../core/localLiteRtLmClient.js';
 import { LlmRole } from '../../telemetry/types.js';
 
@@ -172,16 +173,28 @@ export class NumericalClassifierStrategy implements RoutingStrategy {
           config.getGemini31FlashLiteLaunched(),
           config.getUseCustomToolModel(),
         ]);
-      const selectedModel = resolveClassifierModel(
-        model,
-        modelAlias,
-        useGemini3_1,
-        useGemini3_1FlashLite,
-        useCustomToolModel,
-        config.getHasAccessToPreviewModel?.() ?? true,
-        config,
+      const selectedModel = normalizeModelId(
+        resolveClassifierModel(
+          normalizeModelId(model),
+          modelAlias,
+          useGemini3_1,
+          useGemini3_1FlashLite,
+          useCustomToolModel,
+          config.getHasAccessToPreviewModel?.() ?? true,
+          config,
+        ),
       );
 
+      const service = config.getModelAvailabilityService();
+      const snapshot = service.snapshot(selectedModel);
+
+      if (!snapshot.available) {
+        debugLogger.warn(
+          `[Routing] Numerical classifier selected unavailable model ${selectedModel} (${snapshot.reason}). Bypassing.`,
+        );
+        return null;
+      }
+
       const latencyMs = Date.now() - startTime;
 
       return {