diff --git a/apps/web/src/lib/cloud-agent-next/cloud-agent-client.ts b/apps/web/src/lib/cloud-agent-next/cloud-agent-client.ts
index 971c8ada5d..79eb8325b1 100644
--- a/apps/web/src/lib/cloud-agent-next/cloud-agent-client.ts
+++ b/apps/web/src/lib/cloud-agent-next/cloud-agent-client.ts
@@ -119,6 +119,8 @@ export type AgentMode = string;
 type PrepareSessionSharedFields = {
   mode: AgentMode;
   model: string;
+  /** Optional cheap same-vendor model for title/aux calls (Code Reviewer). */
+  smallModel?: string;
   variant?: string;
   // GitHub-specific params
   githubRepo?: string;
diff --git a/apps/web/src/lib/code-reviews/core/model-selection.test.ts b/apps/web/src/lib/code-reviews/core/model-selection.test.ts
index f6cc0d36a3..8179c3e02d 100644
--- a/apps/web/src/lib/code-reviews/core/model-selection.test.ts
+++ b/apps/web/src/lib/code-reviews/core/model-selection.test.ts
@@ -1,5 +1,10 @@
-import { resolveEffectiveModel, selectedModelFromReviewSources } from './model-selection';
-import type { CodeReviewAgentConfig } from '@kilocode/db/schema-types';
+import {
+  catalogPricesFromStoredModels,
+  resolveCheapSameVendorSmallModel,
+  resolveEffectiveModel,
+  selectedModelFromReviewSources,
+} from './model-selection';
+import type { CodeReviewAgentConfig, StoredModel } from '@kilocode/db/schema-types';
 
 const FALLBACK = 'anthropic/claude-sonnet-4.6';
 
@@ -15,6 +20,15 @@ function baseConfig(
   };
 }
 
+function priced(id: string, prompt: string): StoredModel {
+  return {
+    id,
+    name: id,
+    type: 'language',
+    endpoints: [{ pricing: { prompt, completion: prompt } }],
+  };
+}
+
 describe('resolveEffectiveModel', () => {
   it('uses the global model when there are no overrides', () => {
     const result = resolveEffectiveModel(baseConfig(), 'acme/api', FALLBACK);
@@ -153,3 +167,52 @@ describe('selectedModelFromReviewSources', () => {
     ).toBeNull();
   });
 });
+
+describe('resolveCheapSameVendorSmallModel', () => {
+  const catalog = catalogPricesFromStoredModels({
+    'anthropic/claude-sonnet-4.6': priced('anthropic/claude-sonnet-4.6', '0.000003'),
+    'anthropic/claude-haiku-4.5': priced('anthropic/claude-haiku-4.5', '0.0000008'),
+    'anthropic/claude-opus-4.6': priced('anthropic/claude-opus-4.6', '0.000015'),
+    'openai/gpt-5': priced('openai/gpt-5', '0.00001'),
+    'openai/gpt-5-nano': priced('openai/gpt-5-nano', '0.0000001'),
+  });
+
+  it('picks the cheapest strictly-cheaper same-vendor sibling', () => {
+    expect(resolveCheapSameVendorSmallModel('anthropic/claude-sonnet-4.6', catalog)).toBe(
+      'anthropic/claude-haiku-4.5'
+    );
+  });
+
+  it('does not cross vendors', () => {
+    expect(resolveCheapSameVendorSmallModel('openai/gpt-5', catalog)).toBe('openai/gpt-5-nano');
+  });
+
+  it('leaves managed models unset when no cheaper sibling exists', () => {
+    expect(resolveCheapSameVendorSmallModel('anthropic/claude-haiku-4.5', catalog)).toBeUndefined();
+  });
+
+  it('falls back to the primary for sole-model direct-BYOK vendors', () => {
+    expect(resolveCheapSameVendorSmallModel('neuralwatt/glm-5.2-short', catalog)).toBe(
+      'neuralwatt/glm-5.2-short'
+    );
+  });
+
+  it('does not substitute a Kilo-billed catalog sibling for an uncatalogued BYOK primary', () => {
+    const mixedCatalog = catalogPricesFromStoredModels({
+      'neuralwatt/glm-tiny': priced('neuralwatt/glm-tiny', '0.0000001'),
+    });
+    expect(resolveCheapSameVendorSmallModel('neuralwatt/glm-5.2-short', mixedCatalog)).toBe(
+      'neuralwatt/glm-5.2-short'
+    );
+  });
+
+  it('picks a cheaper BYOK sibling when catalog prices exist', () => {
+    const byokCatalog = catalogPricesFromStoredModels({
+      'neuralwatt/glm-5.2-short': priced('neuralwatt/glm-5.2-short', '0.000002'),
+      'neuralwatt/glm-tiny': priced('neuralwatt/glm-tiny', '0.0000001'),
+    });
+    expect(resolveCheapSameVendorSmallModel('neuralwatt/glm-5.2-short', byokCatalog)).toBe(
+      'neuralwatt/glm-tiny'
+    );
+  });
+});
diff --git a/apps/web/src/lib/code-reviews/core/model-selection.ts b/apps/web/src/lib/code-reviews/core/model-selection.ts
index 503804e503..6fcff65b64 100644
--- a/apps/web/src/lib/code-reviews/core/model-selection.ts
+++ b/apps/web/src/lib/code-reviews/core/model-selection.ts
@@ -12,8 +12,10 @@
  * or Bitbucket). See `RepositoryModelOverrideSchema`.
  */
 
-import type { CodeReviewAgentConfig } from '@kilocode/db/schema-types';
+import type { CodeReviewAgentConfig, StoredModel } from '@kilocode/db/schema-types';
 import { DEFAULT_CODE_REVIEW_MODEL } from './constants';
+import { DirectUserByokInferenceProviderIdSchema } from '@/lib/ai-gateway/providers/openrouter/inference-provider-id';
+import { getOpenRouterModelsMetadataFromDatabase } from '@/lib/ai-gateway/providers/gateway-models-cache';
 
 export type EffectiveModelSelection = {
   modelSlug: string;
@@ -21,6 +23,113 @@ export type EffectiveModelSelection = {
   source: 'repository_override' | 'global';
 };
 
+/** Catalog entry used to pick a cheap same-vendor small model. */
+export type CatalogModelPrice = {
+  id: string;
+  /** USD per input token; null when the catalog has no usable prompt price. */
+  promptPrice: number | null;
+};
+
+const DIRECT_BYOK_VENDORS = new Set<string>(DirectUserByokInferenceProviderIdSchema.options);
+
+export function modelVendorId(modelId: string): string | undefined {
+  const vendor = modelId.split('/')[0]?.trim();
+  return vendor || undefined;
+}
+
+export function isDirectByokVendor(vendor: string | undefined): boolean {
+  return vendor != null && DIRECT_BYOK_VENDORS.has(vendor);
+}
+
+/**
+ * Build catalog price rows from the OpenRouter `StoredModel` map.
+ * Uses the cheapest endpoint prompt price per model.
+ */
+export function catalogPricesFromStoredModels(
+  models: Record<string, StoredModel>
+): CatalogModelPrice[] {
+  return Object.values(models)
+    .filter(model => (model.type ?? 'language') === 'language' && model.endpoints.length > 0)
+    .map(model => {
+      const prices = model.endpoints
+        .map(endpoint =>
+          endpoint.pricing?.prompt != null ? Number.parseFloat(endpoint.pricing.prompt) : Number.NaN
+        )
+        .filter(price => Number.isFinite(price) && price >= 0);
+      return {
+        id: model.id,
+        promptPrice: prices.length > 0 ? Math.min(...prices) : null,
+      };
+    });
+}
+
+/**
+ * Pick a cheap same-vendor model for Code Reviewer title/aux calls.
+ *
+ * - Prefer the lowest-priced same-vendor sibling that is strictly cheaper than the
+ *   primary when the primary's price is known.
+ * - When the primary has no catalog price, pick the cheapest same-vendor sibling;
+ *   direct-BYOK primaries are not in the Kilo-billed catalog, so they fall back to
+ *   the primary instead of a same-prefix catalog entry.
+ * - Direct-BYOK vendors with no cheaper sibling fall back to the primary (user's
+ *   key) so aux calls do not fall through to kilo-auto/small → Gemma on Kilo credits.
+ * - Managed vendors with no cheaper sibling leave small_model unset.
+ */
+export function resolveCheapSameVendorSmallModel(
+  primaryModelId: string,
+  catalog: readonly CatalogModelPrice[]
+): string | undefined {
+  const vendor = modelVendorId(primaryModelId);
+  if (!vendor) return undefined;
+
+  const byok = isDirectByokVendor(vendor);
+  const primaryEntry = catalog.find(entry => entry.id === primaryModelId);
+  const primaryPrice = primaryEntry?.promptPrice ?? undefined;
+
+  // Direct-BYOK models are served from the user's own provider, not the
+  // Kilo-billed OpenRouter catalog. A same-prefix catalog entry therefore does
+  // not represent the same provider, so never substitute one for a BYOK primary.
+  const siblings =
+    byok && primaryEntry == null
+      ? []
+      : catalog.filter(
+          entry =>
+            entry.id !== primaryModelId &&
+            modelVendorId(entry.id) === vendor &&
+            entry.promptPrice != null &&
+            Number.isFinite(entry.promptPrice)
+        );
+
+  const cheaper =
+    primaryPrice != null && Number.isFinite(primaryPrice)
+      ? siblings.filter(entry => (entry.promptPrice as number) < primaryPrice)
+      : siblings;
+
+  if (cheaper.length === 0) {
+    return byok ? primaryModelId : undefined;
+  }
+
+  cheaper.sort((a, b) => {
+    const priceDelta = (a.promptPrice as number) - (b.promptPrice as number);
+    return priceDelta !== 0 ? priceDelta : a.id.localeCompare(b.id);
+  });
+  return cheaper[0]?.id;
+}
+
+/**
+ * Resolve the Code Reviewer aux/title small model for a primary review model.
+ * Soft-fails to BYOK-primary or unset when the gateway catalog cannot be loaded.
+ */
+export async function resolveReviewSmallModel(primaryModelId: string): Promise<string | undefined> {
+  const vendor = modelVendorId(primaryModelId);
+  try {
+    const models = await getOpenRouterModelsMetadataFromDatabase();
+    return resolveCheapSameVendorSmallModel(primaryModelId, catalogPricesFromStoredModels(models));
+  } catch {
+    return isDirectByokVendor(vendor) ? primaryModelId : undefined;
+  }
+}
+
 /**
  * Resolve the effective model for a review's repository.
  *
diff --git a/apps/web/src/lib/code-reviews/triggers/prepare-review-payload.test.ts b/apps/web/src/lib/code-reviews/triggers/prepare-review-payload.test.ts
index 2f85c05e2b..d0db527bd2 100644
--- a/apps/web/src/lib/code-reviews/triggers/prepare-review-payload.test.ts
+++ b/apps/web/src/lib/code-reviews/triggers/prepare-review-payload.test.ts
@@ -15,9 +15,11 @@ const mockFindPreviousCompletedReview = jest.fn();
 const mockUpdatePreviousReviewSummary = jest.fn();
 const mockUpdateRepositoryReviewInstructionsMetadata = jest.fn();
 const mockGenerateReviewPrompt = jest.fn();
+const mockResolveReviewSmallModel = jest.fn();
 
 import type { CodeReviewAgentConfig } from '@/lib/agent-config/core/types';
 import type * as CodeReviewsDb from '@/lib/code-reviews/db/code-reviews';
+import type * as ModelSelection from '@/lib/code-reviews/core/model-selection';
 
 jest.mock('@/lib/integrations/platforms/github/adapter', () => ({
   generateGitHubInstallationToken: (...args: unknown[]) =>
@@ -62,6 +64,16 @@ jest.mock('@sentry/nextjs', () => ({
   captureException: jest.fn(),
 }));
 
+jest.mock('@/lib/code-reviews/core/model-selection', () => {
+  const actual = jest.requireActual<typeof ModelSelection>(
+    '@/lib/code-reviews/core/model-selection'
+  );
+  return {
+    ...actual,
+    resolveReviewSmallModel: (...args: unknown[]) => mockResolveReviewSmallModel(...args),
+  };
+});
+
 import { db } from '@/lib/drizzle';
 import { insertTestUser } from '@/tests/helpers/user.helper';
 import {
@@ -249,6 +261,8 @@ describe('prepareReviewPayload', () => {
     mockUpdatePreviousReviewSummary.mockReset();
     mockUpdateRepositoryReviewInstructionsMetadata.mockReset();
     mockGenerateReviewPrompt.mockReset();
+    mockResolveReviewSmallModel.mockReset();
+    mockResolveReviewSmallModel.mockResolvedValue(undefined);
   });
 
   afterAll(async () => {
@@ -979,4 +993,39 @@ describe('prepareReviewPayload', () => {
       upstreamBranch: 'refs/pull/1235/head',
     });
   });
+
+  it('includes resolved smallModel on GitHub session input when present', async () => {
+    const [review] = await db
+      .insert(cloud_agent_code_reviews)
+      .values(defineReview(testUser.id, integration.id))
+      .returning();
+    mockResolveReviewSmallModel.mockResolvedValueOnce('anthropic/claude-haiku-4.5');
+
+    const payload = await prepareReviewPayload({
+      reviewId: review.id,
+      owner: { type: 'user', id: testUser.id, userId: testUser.id },
+      agentConfig: { config: baseAgentConfig },
+      platform: 'github',
+    });
+
+    expect(mockResolveReviewSmallModel).toHaveBeenCalledWith('test-model');
+    expect(payload.sessionInput.smallModel).toBe('anthropic/claude-haiku-4.5');
+  });
+
+  it('omits smallModel from session input when no cheap sibling is resolved', async () => {
+    const [review] = await db
+      .insert(cloud_agent_code_reviews)
+      .values(defineReview(testUser.id, integration.id))
+      .returning();
+
+    const payload = await prepareReviewPayload({
+      reviewId: review.id,
+      owner: { type: 'user', id: testUser.id, userId: testUser.id },
+      agentConfig: { config: baseAgentConfig },
+      platform: 'github',
+    });
+
+    expect(mockResolveReviewSmallModel).toHaveBeenCalledWith('test-model');
+    expect(payload.sessionInput).not.toHaveProperty('smallModel');
+  });
 });
diff --git a/apps/web/src/lib/code-reviews/triggers/prepare-review-payload.ts b/apps/web/src/lib/code-reviews/triggers/prepare-review-payload.ts
index 65ee241b82..4a36b53eeb 100644
--- a/apps/web/src/lib/code-reviews/triggers/prepare-review-payload.ts
+++ b/apps/web/src/lib/code-reviews/triggers/prepare-review-payload.ts
@@ -60,6 +60,7 @@ import {
   type ReviewScope,
 } from '../db/code-reviews';
 import { DEFAULT_CODE_REVIEW_MODEL, DEFAULT_CODE_REVIEW_MODE } from '../core/constants';
+import { resolveReviewSmallModel } from '../core/model-selection';
 import type { Owner } from '../core';
 import { generateReviewPrompt } from '../prompts/generate-prompt';
 import type { CodeReviewAgentConfig } from '@/lib/agent-config/core/types';
@@ -113,6 +114,11 @@ export type SessionInput = {
   prompt: string;
   mode: 'code';
   model: string;
+  /**
+   * Optional cheap same-vendor model for kilo title/aux calls. When omitted, the
+   * CLI falls through to its default small-model list (kilo-auto/small → Gemma).
+   */
+  smallModel?: string;
   /** Thinking effort variant name (e.g. "high", "max") — undefined means model default */
   variant?: string;
   upstreamBranch: string;
@@ -302,12 +308,14 @@ export async function prepareReviewPayload(
         // Single source for the standard reviewer's model so the session input and the
         // forward-shaped `reviewAgents[0]` can never drift apart.
         const standardModel = config.model_slug || DEFAULT_CODE_REVIEW_MODEL;
+        const smallModel = await resolveReviewSmallModel(standardModel);
         const sessionInput: SessionInput = {
           gitUrl: `https://bitbucket.org/${workspaceSlug.data}/${repositorySlug.data}.git`,
           kilocodeOrganizationId: owner.id,
           prompt,
           mode: DEFAULT_CODE_REVIEW_MODE as 'code',
           model: standardModel,
+          ...(smallModel ? { smallModel } : {}),
           variant: config.thinking_effort ?? undefined,
           upstreamBranch: review.head_ref,
           platform: PLATFORM.BITBUCKET,
@@ -758,6 +766,7 @@ export async function prepareReviewPayload(
     // Single source for the standard reviewer's model so the session input and the
     // forward-shaped `reviewAgents[0]` can never drift apart.
     const standardModel = config.model_slug || DEFAULT_CODE_REVIEW_MODEL;
+    const smallModel = await resolveReviewSmallModel(standardModel);
     const gateThreshold = config.gate_threshold ?? 'off';
     const githubCheckoutRef = getGitHubPullRequestCheckoutRef(review.pr_number);
     const sessionInput: SessionInput =
@@ -771,6 +780,7 @@ export async function prepareReviewPayload(
             prompt,
             mode: DEFAULT_CODE_REVIEW_MODE as 'code',
             model: standardModel,
+            ...(smallModel ? { smallModel } : {}),
             variant,
             upstreamBranch: review.head_ref,
           }
@@ -784,6 +794,7 @@ export async function prepareReviewPayload(
               prompt,
               mode: DEFAULT_CODE_REVIEW_MODE as 'code',
               model: standardModel,
+              ...(smallModel ? { smallModel } : {}),
               variant,
               upstreamBranch: review.head_ref,
               ...(gateThreshold !== 'off' ? { gateThreshold } : {}),
@@ -797,6 +808,7 @@ export async function prepareReviewPayload(
               prompt,
               mode: DEFAULT_CODE_REVIEW_MODE as 'code',
               model: standardModel,
+              ...(smallModel ? { smallModel } : {}),
               variant,
               upstreamBranch: githubCheckoutRef,
               ...(gateThreshold !== 'off' ? { gateThreshold } : {}),
diff --git a/packages/worker-utils/src/cloud-agent-next-client.ts b/packages/worker-utils/src/cloud-agent-next-client.ts
index 3d15f59722..d2890aceeb 100644
--- a/packages/worker-utils/src/cloud-agent-next-client.ts
+++ b/packages/worker-utils/src/cloud-agent-next-client.ts
@@ -44,6 +44,8 @@ export type CloudAgentPrepareSessionInput = {
   prompt: string;
   mode: string;
   model: string;
+  /** Optional cheap same-vendor model for title/aux calls (Code Reviewer). */
+  smallModel?: string;
   variant?: string;
   githubRepo?: string;
   githubToken?: string;
diff --git a/services/cloud-agent-next/src/execution/types.ts b/services/cloud-agent-next/src/execution/types.ts
index 8fd5fc64ca..fe146e1005 100644
--- a/services/cloud-agent-next/src/execution/types.ts
+++ b/services/cloud-agent-next/src/execution/types.ts
@@ -85,6 +85,8 @@ export function renderExecutionTurnContent(turn: AcceptedExecutionTurn): string
 export type ModelChoice = {
   model: string;
   variant?: string;
+  /** Optional cheap same-vendor model for title/aux calls (Code Reviewer). */
+  smallModel?: string;
 };
 
 /** Fully resolved agent selection. */
diff --git a/services/cloud-agent-next/src/persistence/CloudAgentSession.ts b/services/cloud-agent-next/src/persistence/CloudAgentSession.ts
index b567e6aae9..16b66d701c 100644
--- a/services/cloud-agent-next/src/persistence/CloudAgentSession.ts
+++ b/services/cloud-agent-next/src/persistence/CloudAgentSession.ts
@@ -447,6 +447,7 @@ function isSameInitialAdmissionConfiguration(
       JSON.stringify(input.workspace?.sandboxRoute) &&
     metadata.agent?.mode === input.agent.mode &&
     metadata.agent.model === input.agent.model &&
+    metadata.agent.smallModel === input.agent.smallModel &&
     metadata.agent.variant === input.agent.variant &&
     metadata.finalization?.autoCommit === input.finalization?.autoCommit &&
     metadata.finalization?.condenseOnComplete === input.finalization?.condenseOnComplete
@@ -2551,6 +2552,7 @@ export class CloudAgentSession extends DurableObject<WorkerEnv> {
       agent: {
         mode: input.agent.mode,
         model: input.agent.model,
+        smallModel: input.agent.smallModel,
         variant: input.agent.variant,
         appendSystemPrompt: input.agent.appendSystemPrompt,
       },
@@ -3766,6 +3768,7 @@ export class CloudAgentSession extends DurableObject<WorkerEnv> {
       agent: {
         mode: metadata.agent.mode,
         model: metadata.agent.model,
+        smallModel: metadata.agent.smallModel,
         variant: metadata.agent.variant,
       },
       finalization: {
diff --git a/services/cloud-agent-next/src/persistence/session-metadata.test.ts b/services/cloud-agent-next/src/persistence/session-metadata.test.ts
index e9c5036a3b..af81c61fec 100644
--- a/services/cloud-agent-next/src/persistence/session-metadata.test.ts
+++ b/services/cloud-agent-next/src/persistence/session-metadata.test.ts
@@ -272,6 +272,7 @@ describe('session metadata boundary', () => {
       agent: {
         mode: 'reviewer',
         model: 'kilo/gpt-5',
+        smallModel: 'kilo/anthropic/claude-haiku-4.5',
         variant: 'thinking',
         appendSystemPrompt: 'Extra context',
       },
diff --git a/services/cloud-agent-next/src/persistence/session-metadata.ts b/services/cloud-agent-next/src/persistence/session-metadata.ts
index 63f1f55353..b14838b0eb 100644
--- a/services/cloud-agent-next/src/persistence/session-metadata.ts
+++ b/services/cloud-agent-next/src/persistence/session-metadata.ts
@@ -193,6 +193,7 @@ const MetadataAgentSchema = z
   .object({
     mode: z.string().optional(),
     model: z.string().optional(),
+    smallModel: z.string().optional(),
     variant: z
       .string()
       .max(50)
diff --git a/services/cloud-agent-next/src/router/handlers/session-prepare.ts b/services/cloud-agent-next/src/router/handlers/session-prepare.ts
index 007ab40850..6fa6ae18fe 100644
--- a/services/cloud-agent-next/src/router/handlers/session-prepare.ts
+++ b/services/cloud-agent-next/src/router/handlers/session-prepare.ts
@@ -270,6 +270,7 @@ export function prepareInputToSessionCreateRequest(input: PrepareInput): Session
     agent: {
       mode: input.mode,
       model: input.model,
+      ...(input.smallModel ? { smallModel: input.smallModel } : {}),
       variant: input.variant,
     },
     repository,
diff --git a/services/cloud-agent-next/src/router/schemas.ts b/services/cloud-agent-next/src/router/schemas.ts
index 35eb78c7db..61bd23d7bc 100644
--- a/services/cloud-agent-next/src/router/schemas.ts
+++ b/services/cloud-agent-next/src/router/schemas.ts
@@ -436,6 +436,11 @@ const PrepareSessionSharedFields = {
     'Kilo Code execution mode (built-in or custom slug from runtimeAgents)'
   ),
   model: modelIdSchema.describe('AI model to use'),
+  smallModel: modelIdSchema
+    .optional()
+    .describe(
+      'Optional cheap same-vendor model for title/aux calls (Code Reviewer). When omitted, CLI defaults apply.'
+    ),
   variant: z
     .string()
     .max(50)
diff --git a/services/cloud-agent-next/src/session-service.test.ts b/services/cloud-agent-next/src/session-service.test.ts
index dd988ba3da..ff2fd82bb6 100644
--- a/services/cloud-agent-next/src/session-service.test.ts
+++ b/services/cloud-agent-next/src/session-service.test.ts
@@ -178,6 +178,32 @@ describe('SessionService.buildRuntimeEnv', () => {
     expect(runtimeEnv.GIT_CONFIG_GLOBAL).toBeUndefined();
     expect(runtimeEnv.GIT_CONFIG_NOSYSTEM).toBeUndefined();
   });
+
+  it('materializes code-review small_model from the smallModel option', () => {
+    const service = new SessionService();
+    const context = service.buildContext({
+      sandboxId: 'usr-test',
+      userId: 'user_test',
+      sessionId: 'agent_test',
+    });
+    const smallModel = 'anthropic/claude-haiku-4.5';
+
+    const runtimeEnv = service.buildRuntimeEnv({
+      context,
+      env: createEnv(),
+      kiloCapability: 'kilo-token',
+      kilocodeModel: 'anthropic/claude-sonnet-4.6',
+      smallModel,
+      createdOnPlatform: 'code-review',
+    });
+
+    const config = JSON.parse(runtimeEnv.KILO_CONFIG_CONTENT) as {
+      small_model?: string;
+      agent?: { title?: { model?: string } };
+    };
+    expect(config.small_model).toBe(`kilo/${smallModel}`);
+    expect(config.agent?.title?.model).toBe(`kilo/${smallModel}`);
+  });
 });
 
 describe('code-review command guard policy', () => {
@@ -2444,6 +2470,140 @@ describe('SessionService.buildWrapperSessionReadyAndPromptRequests', () => {
     });
   }
 
+  it('pins code-review KILO_CONFIG small_model to the dispatched smallModel', async () => {
+    const service = new SessionService();
+    const env = createEnv();
+    env.WORKER_URL = 'https://cloud-agent.example.com';
+    const primaryModel = 'anthropic/claude-sonnet-4.6';
+    const smallModel = 'anthropic/claude-haiku-4.5';
+
+    const build = (args: {
+      messageId: string;
+      wrapperRunId: string;
+      metadata: CloudAgentSessionState;
+      agent: { mode: 'code'; model: string; smallModel?: string };
+    }) =>
+      service.buildWrapperSessionReadyAndPromptRequests({
+        env,
+        plan: {
+          scope: { sessionId: 'agent_test', userId: 'user_test' },
+          turn: { type: 'prompt', messageId: args.messageId, prompt: 'Review the PR' },
+          agent: args.agent,
+          workspace: {
+            sandboxId: args.metadata.workspace?.sandboxId ?? 'ses-abcdef',
+            metadata: args.metadata,
+          },
+          wrapper: {
+            fence: {
+              wrapperRunId: args.wrapperRunId,
+              wrapperGeneration: 1,
+              wrapperConnectionId: `conn_${args.wrapperRunId}`,
+            },
+          },
+        } satisfies FencedWrapperDispatchRequest,
+      });
+
+    const readConfig = (result: Awaited<ReturnType<typeof build>>) =>
+      JSON.parse(result.readyRequest.materialized.env.KILO_CONFIG_CONTENT) as {
+        model?: string;
+        small_model?: string;
+        agent?: { title?: { model?: string } };
+      };
+
+    const codeReview = readConfig(
+      await build({
+        messageId: 'msg_018f1e2d3c4bSmallModelAAAA',
+        wrapperRunId: 'wr_small_model',
+        metadata: createMetadata({ createdOnPlatform: 'code-review' }),
+        agent: { mode: 'code', model: primaryModel, smallModel },
+      })
+    );
+    expect(codeReview.model).toBe(`kilo/${primaryModel}`);
+    expect(codeReview.small_model).toBe(`kilo/${smallModel}`);
+    expect(codeReview.agent?.title?.model).toBe(`kilo/${smallModel}`);
+
+    const unset = readConfig(
+      await build({
+        messageId: 'msg_018f1e2d3c4bSmallModelNone',
+        wrapperRunId: 'wr_small_model_unset',
+        metadata: createMetadata({ createdOnPlatform: 'code-review' }),
+        agent: { mode: 'code', model: primaryModel },
+      })
+    );
+    expect(unset.model).toBe(`kilo/${primaryModel}`);
+    expect(unset.small_model).toBeUndefined();
+    expect(unset.agent?.title).toBeUndefined();
+
+    const web = readConfig(
+      await build({
+        messageId: 'msg_018f1e2d3c4bSmallModelBBBB',
+        wrapperRunId: 'wr_small_model_web',
+        metadata: createMetadata({ createdOnPlatform: 'cloud-agent-web' }),
+        agent: { mode: 'code', model: primaryModel, smallModel },
+      })
+    );
+    expect(web.model).toBe(`kilo/${primaryModel}`);
+    expect(web.small_model).toBeUndefined();
+    expect(web.agent?.title).toBeUndefined();
+
+    // Production code-review sessions persist smallModel in metadata; the plan
+    // agent does not carry it, so this exercises the metadata fallback.
+    const baseMetadata = createMetadata({ createdOnPlatform: 'code-review' });
+    const fromMetadata = readConfig(
+      await build({
+        messageId: 'msg_018f1e2d3c4bSmallModelCCCC',
+        wrapperRunId: 'wr_small_model_metadata',
+        metadata: { ...baseMetadata, agent: { ...baseMetadata.agent, smallModel } },
+        agent: { mode: 'code', model: primaryModel },
+      })
+    );
+    expect(fromMetadata.model).toBe(`kilo/${primaryModel}`);
+    expect(fromMetadata.small_model).toBe(`kilo/${smallModel}`);
+    expect(fromMetadata.agent?.title?.model).toBe(`kilo/${smallModel}`);
+  });
+
+  it('sets Bitbucket code-review small_model without injecting agent.title', async () => {
+    const service = new SessionService();
+    const env = createEnv();
+    env.WORKER_URL = 'https://cloud-agent.example.com';
+    const primaryModel = 'anthropic/claude-sonnet-4.6';
+    const smallModel = 'anthropic/claude-haiku-4.5';
+    const metadata = createBitbucketMetadata(true);
+
+    const result = await service.buildWrapperSessionReadyAndPromptRequests({
+      env,
+      plan: {
+        scope: { sessionId: 'agent_test', userId: 'user_test' },
+        turn: {
+          type: 'prompt',
+          messageId: 'msg_018f1e2d3c4bSmallModelBBBB',
+          prompt: 'Review the PR',
+        },
+        agent: { mode: 'code', model: primaryModel, smallModel },
+        workspace: {
+          sandboxId: metadata.workspace?.sandboxId ?? 'ses-abcdef',
+          metadata,
+        },
+        wrapper: {
+          fence: {
+            wrapperRunId: 'wr_bb_small_model',
+            wrapperGeneration: 1,
+            wrapperConnectionId: 'conn_bb_small_model',
+          },
+        },
+      } satisfies FencedWrapperDispatchRequest,
+    });
+
+    const config = JSON.parse(result.readyRequest.materialized.env.KILO_CONFIG_CONTENT) as {
+      model?: string;
+      small_model?: string;
+      agent?: { title?: { model?: string } };
+    };
+    expect(config.model).toBe(`kilo/${primaryModel}`);
+    expect(config.small_model).toBe(`kilo/${smallModel}`);
+    expect(config.agent?.title).toBeUndefined();
+  });
+
   it('uses the persisted shared checkout when constructing wrapper requests', async () => {
     const worktreeId = 'worktree_420ae020-e3c4-4e67-878b-66672c3d997e';
     const workspacePath = `/workspace/user_test/worktrees/${worktreeId}`;
diff --git a/services/cloud-agent-next/src/session-service.ts b/services/cloud-agent-next/src/session-service.ts
index 3cb68c4038..bfa49efde2 100644
--- a/services/cloud-agent-next/src/session-service.ts
+++ b/services/cloud-agent-next/src/session-service.ts
@@ -1282,6 +1282,7 @@ export class SessionService {
       kiloProviderBaseUrl: opts.kiloProviderBaseUrl,
       kiloSessionIngestBaseUrl: opts.kiloSessionIngestBaseUrl,
       kilocodeModel: opts.kilocodeModel,
+      smallModel: opts.smallModel,
       originalOrgId: opts.originalOrgId,
       githubToken: context.githubToken,
       githubRepo: context.githubRepo,
@@ -1310,6 +1311,7 @@ export class SessionService {
       kiloProviderBaseUrl,
       kiloSessionIngestBaseUrl,
       kilocodeModel,
+      smallModel,
       originalOrgId,
       githubToken,
       githubRepo,
@@ -1535,6 +1537,25 @@ export class SessionService {
         agentCount: runtimeAgents.length,
       });
     }
+    // Code Review sessions: pin small_model (and title agent when allowed) to the
+    // cheap same-vendor model selected at dispatch time. Leaving this unset falls
+    // through to kilo-auto/small → Gemma. See https://github.com/Kilo-Org/cloud/issues/4268
+    const normalizedSmallModel =
+      smallModel && smallModel.trim() ? normalizeKilocodeModel(smallModel) : undefined;
+    if (createdOnPlatform === 'code-review' && normalizedSmallModel) {
+      configContent.small_model = normalizedSmallModel;
+      // Match MCP/runtimeAgents: do not inject agent.title on Bitbucket code-review
+      // sessions (title falls through to getSmallModel() via small_model anyway).
+      if (!bitbucketInputPath) {
+        const existingTitle =
+          agentConfig.title != null &&
+          typeof agentConfig.title === 'object' &&
+          !Array.isArray(agentConfig.title)
+            ? (agentConfig.title as Record<string, unknown>)
+            : {};
+        agentConfig.title = { ...existingTitle, model: normalizedSmallModel };
+      }
+    }
     if (Object.keys(agentConfig).length > 0) {
       configContent.agent = agentConfig;
     }
@@ -1669,6 +1690,7 @@ export class SessionService {
       kiloCapability,
       kiloProviderBaseUrl,
       kilocodeModel,
+      smallModel,
       originalOrgId,
       createdOnPlatform,
       callbackTarget,
@@ -1688,6 +1710,7 @@ export class SessionService {
       kiloCapability,
       kiloProviderBaseUrl,
       kilocodeModel,
+      smallModel,
       originalOrgId,
       createdOnPlatform,
       callbackTarget,
@@ -2121,6 +2144,7 @@ export class SessionService {
       kiloProviderBaseUrl,
       kiloSessionIngestBaseUrl,
       kilocodeModel: agent.model,
+      smallModel: metadata.agent?.smallModel ?? agent.smallModel,
       originalOrgId: orgId,
       githubToken: resolvedTokens.githubToken,
       githubRepo: github?.repo,
@@ -2400,6 +2424,7 @@ export class SessionService {
       kiloProviderBaseUrl,
       kiloSessionIngestBaseUrl,
       kilocodeModel: options.kilocodeModel,
+      smallModel: metadata.agent?.smallModel,
       originalOrgId: orgId,
       createdOnPlatform: metadata.identity.createdOnPlatform,
       callbackTarget: metadata.callback?.target,
@@ -2620,6 +2645,7 @@ export class SessionService {
       kiloProviderBaseUrl,
       kiloSessionIngestBaseUrl,
       kilocodeModel,
+      smallModel: metadata.agent?.smallModel,
       originalOrgId: orgId,
       createdOnPlatform: metadata.identity.createdOnPlatform,
       callbackTarget: metadata.callback?.target,
@@ -3032,6 +3058,7 @@ export type GetOrCreateSessionOptions = {
   kiloProviderBaseUrl?: string;
   kiloSessionIngestBaseUrl?: string;
   kilocodeModel?: string;
+  smallModel?: string;
   originalOrgId?: string;
   createdOnPlatform?: string;
   callbackTarget?: NonNullable<CloudAgentSessionState['callback']>['target'];
@@ -3060,6 +3087,8 @@ type GetSaferEnvVarsOptions = {
   kiloProviderBaseUrl?: string;
   kiloSessionIngestBaseUrl?: string;
   kilocodeModel?: string;
+  /** Optional cheap same-vendor model for Code Reviewer title/aux calls. */
+  smallModel?: string;
   originalOrgId?: string;
   githubToken?: string;
   githubRepo?: string;
diff --git a/services/cloud-agent-next/src/session/session-prepare.test.ts b/services/cloud-agent-next/src/session/session-prepare.test.ts
index d6e7f8adb2..9b086e67fd 100644
--- a/services/cloud-agent-next/src/session/session-prepare.test.ts
+++ b/services/cloud-agent-next/src/session/session-prepare.test.ts
@@ -4125,4 +4125,36 @@ describe('prepareInputToSessionCreateRequest clone mapping', () => {
 
     expect(request.runtime).toEqual({ sandboxAllocation: 'isolated-standard' });
   });
+
+  it('maps a Code Reviewer smallModel into the grouped agent selection', () => {
+    const request = prepareInputToSessionCreateRequest({
+      prompt: 'Review the PR',
+      mode: 'code',
+      model: 'anthropic/claude-sonnet-4.6',
+      smallModel: 'anthropic/claude-haiku-4.5',
+      githubRepo: 'acme/repo',
+      shallow: false,
+      devcontainer: false,
+    });
+
+    expect(request.agent).toEqual({
+      mode: 'code',
+      model: 'anthropic/claude-sonnet-4.6',
+      smallModel: 'anthropic/claude-haiku-4.5',
+      variant: undefined,
+    });
+  });
+
+  it('omits smallModel from the grouped agent selection when absent', () => {
+    const request = prepareInputToSessionCreateRequest({
+      prompt: 'Review the PR',
+      mode: 'code',
+      model: 'anthropic/claude-sonnet-4.6',
+      githubRepo: 'acme/repo',
+      shallow: false,
+      devcontainer: false,
+    });
+
+    expect(request.agent).not.toHaveProperty('smallModel');
+  });
 });
diff --git a/services/cloud-agent-next/src/session/session-registration.ts b/services/cloud-agent-next/src/session/session-registration.ts
index ec9a6671b3..c646bd32bc 100644
--- a/services/cloud-agent-next/src/session/session-registration.ts
+++ b/services/cloud-agent-next/src/session/session-registration.ts
@@ -1380,6 +1380,9 @@ export async function sessionCreateIntentFingerprint(
         mode: input.agent.mode,
         model: input.agent.model,
         variant: input.agent.variant || undefined,
+        // Cheap same-vendor aux model is immutable create input: it changes the
+        // materialized session env, so a changed value must not replay.
+        smallModel: input.agent.smallModel || undefined,
         // The DO stores the effective appended system prompt under `agent`;
         // it is immutable create input, so a changed system prompt must never
         // replay or reconcile a prior session.
diff --git a/services/cloud-agent-next/test/helpers/session-setup.ts b/services/cloud-agent-next/test/helpers/session-setup.ts
index 7f9821123e..ed70b6bbe0 100644
--- a/services/cloud-agent-next/test/helpers/session-setup.ts
+++ b/services/cloud-agent-next/test/helpers/session-setup.ts
@@ -33,6 +33,7 @@ type TestRegisterSessionInput = {
   prompt: string;
   mode: AgentMode;
   model: string;
+  smallModel?: string;
   variant?: string;
   kiloSessionId?: string;
   kilocodeToken?: string;
@@ -147,6 +148,7 @@ export function groupedRegisterSessionInput(input: TestRegisterSessionInput): Re
     agent: {
       mode: input.mode,
       model: input.model,
+      smallModel: input.smallModel,
       variant: input.variant,
       appendSystemPrompt: input.appendSystemPrompt,
     },
diff --git a/services/cloud-agent-next/test/integration/session/code-review-small-model.test.ts b/services/cloud-agent-next/test/integration/session/code-review-small-model.test.ts
new file mode 100644
index 0000000000..3860a0b566
--- /dev/null
+++ b/services/cloud-agent-next/test/integration/session/code-review-small-model.test.ts
@@ -0,0 +1,53 @@
+import { env, runInDurableObject } from 'cloudflare:test';
+import { describe, expect, it } from 'vitest';
+import { registerReadySession } from '../../helpers/session-setup.js';
+
+describe('code-review small model metadata', () => {
+  it('persists agent.smallModel through registerSession', async () => {
+    const userId = 'user_small_model';
+    const sessionId = 'agent_small_model';
+    const stub = env.CLOUD_AGENT_SESSION.get(
+      env.CLOUD_AGENT_SESSION.idFromName(`${userId}:${sessionId}`)
+    );
+
+    await runInDurableObject(stub, async instance => {
+      await registerReadySession(instance, {
+        sessionId,
+        userId,
+        prompt: 'Review the PR',
+        mode: 'code',
+        model: 'anthropic/claude-sonnet-4.6',
+        smallModel: 'anthropic/claude-haiku-4.5',
+        createdOnPlatform: 'code-review',
+        kilocodeToken: 'token-small-model',
+      });
+
+      const metadata = await instance.getMetadata();
+      expect(metadata?.agent?.smallModel).toBe('anthropic/claude-haiku-4.5');
+      expect(metadata?.agent?.model).toBe('anthropic/claude-sonnet-4.6');
+    });
+  });
+
+  it('omits agent.smallModel when the caller does not provide one', async () => {
+    const userId = 'user_small_model_none';
+    const sessionId = 'agent_small_model_none';
+    const stub = env.CLOUD_AGENT_SESSION.get(
+      env.CLOUD_AGENT_SESSION.idFromName(`${userId}:${sessionId}`)
+    );
+
+    await runInDurableObject(stub, async instance => {
+      await registerReadySession(instance, {
+        sessionId,
+        userId,
+        prompt: 'Do the work',
+        mode: 'code',
+        model: 'anthropic/claude-sonnet-4.6',
+        createdOnPlatform: 'cloud-agent-web',
+        kilocodeToken: 'token-small-model-none',
+      });
+
+      const metadata = await instance.getMetadata();
+      expect(metadata?.agent?.smallModel).toBeUndefined();
+    });
+  });
+});
diff --git a/services/code-review-infra/src/types.ts b/services/code-review-infra/src/types.ts
index 17b42630eb..0f03444246 100644
--- a/services/code-review-infra/src/types.ts
+++ b/services/code-review-infra/src/types.ts
@@ -26,6 +26,8 @@ export interface SessionInput {
   prompt: string;
   mode: 'code';
   model: string;
+  /** Optional cheap same-vendor model for kilo title/aux calls. */
+  smallModel?: string;
   /** Thinking effort variant name (e.g. "high", "max") — undefined means model default */
   variant?: string;
   upstreamBranch: string;