Update model recommendations and defaults

apps/web/src/lib/ai-gateway/auto-model/index.ts

@@ -9,7 +9,6 @@ import {
   type OpenCodeSettings,
   type Verbosity,
 } from '@kilocode/db/schema-types';
-import { KIMI_CURRENT_MODEL_ID } from '@/lib/ai-gateway/providers/moonshotai';
 import {
   gemma_4_26b_a4b_it_free_model,
   GEMMA_4_26B_A4B_IT_ID,
@@ -85,17 +84,6 @@ export const FRONTIER_MODE_TO_MODEL: Record<Mode, ResolvedAutoModel> = {
   code: SONNET_FRONTIER,
 };
 
-// INVARIANT: the efficient static fallback must remain image-capable.
-// The capability-aware routing filter relies on this guarantee to make
-// image requests succeed even when no benchmark candidate is capable.
-// Whoever changes this model constant must re-verify image support
-// (via live OpenRouter data or the `model_stats` table) before
-// swapping it — do not assume parity with the prior value.
-export const BALANCED_FALLBACK_MODEL: ResolvedAutoModel = {
-  model: KIMI_CURRENT_MODEL_ID,
-  reasoning: { enabled: true },
-};
-
 const UNKNOWN_PRICING: AutoModelPricing = {
   prompt: '-1',
   completion: '-1',

apps/web/src/lib/ai-gateway/auto-model/resolution.test.ts

@@ -6,14 +6,14 @@ jest.mock('@/lib/ai-gateway/providers/gateway-models-cache', () => ({
 
 import { resolveAutoModel } from './resolution';
 import {
-  BALANCED_FALLBACK_MODEL,
   FRONTIER_MODE_TO_MODEL,
   KILO_AUTO_BALANCED_MODEL,
   KILO_AUTO_EFFICIENT_MODEL,
   KILO_AUTO_FREE_MODEL,
   ORG_AUTO_MODEL,
 } from '@/lib/ai-gateway/auto-model';
 import type { AutoRoutingDecision } from '@kilocode/auto-routing-contracts';
+import { PRIMARY_DEFAULT_MODEL } from '@/lib/ai-gateway/models';
 
 const baseParams = {
   model: KILO_AUTO_EFFICIENT_MODEL.id,
@@ -26,6 +26,7 @@ const baseParams = {
 
 const nullUserPromise = Promise.resolve(null);
 const zeroBalancePromise = Promise.resolve(0);
+const primaryDefaultFallback = { model: PRIMARY_DEFAULT_MODEL };
 
 const sampleDecision: AutoRoutingDecision = {
   model: 'anthropic/claude-haiku-4',
@@ -100,38 +101,38 @@ describe('resolveAutoModel — kilo-auto/efficient branch', () => {
     expect(result).toEqual({ kind: 'ok', resolved: { model: 'anthropic/claude-haiku-4' } });
   });
 
-  it('falls back to BALANCED_FALLBACK_MODEL when no thunk is provided and apiKind=responses', async () => {
+  it('falls back to PRIMARY_DEFAULT_MODEL when no thunk is provided and apiKind=responses', async () => {
     const result = await resolveAutoModel(
       { ...baseParams, apiKind: 'responses' },
       nullUserPromise,
       zeroBalancePromise
     );
 
-    expect(result).toEqual({ kind: 'ok', resolved: BALANCED_FALLBACK_MODEL });
+    expect(result).toEqual({ kind: 'ok', resolved: primaryDefaultFallback });
   });
 
-  it('falls back to BALANCED_FALLBACK_MODEL when no thunk is provided and apiKind=messages', async () => {
+  it('falls back to PRIMARY_DEFAULT_MODEL when no thunk is provided and apiKind=messages', async () => {
     const result = await resolveAutoModel(
       { ...baseParams, apiKind: 'messages' },
       nullUserPromise,
       zeroBalancePromise
     );
 
-    expect(result).toEqual({ kind: 'ok', resolved: BALANCED_FALLBACK_MODEL });
+    expect(result).toEqual({ kind: 'ok', resolved: primaryDefaultFallback });
   });
 
-  it('falls back to BALANCED_FALLBACK_MODEL when no thunk is provided and apiKind=chat_completions', async () => {
+  it('falls back to PRIMARY_DEFAULT_MODEL when no thunk is provided and apiKind=chat_completions', async () => {
     const result = await resolveAutoModel(
       { ...baseParams, apiKind: 'chat_completions' },
       nullUserPromise,
       zeroBalancePromise
     );
 
-    expect(result).toEqual({ kind: 'ok', resolved: BALANCED_FALLBACK_MODEL });
+    expect(result).toEqual({ kind: 'ok', resolved: primaryDefaultFallback });
   });
 
   it.each([KILO_AUTO_BALANCED_MODEL.id, KILO_AUTO_EFFICIENT_MODEL.id])(
-    'falls back to Kimi K3 with reasoning for %s when the worker returns no decision',
+    'falls back to PRIMARY_DEFAULT_MODEL for %s when the worker returns no decision',
     async model => {
       const result = await resolveAutoModel(
         {
@@ -144,14 +145,11 @@ describe('resolveAutoModel — kilo-auto/efficient branch', () => {
         zeroBalancePromise
       );
 
-      expect(result).toEqual({
-        kind: 'ok',
-        resolved: { model: 'moonshotai/kimi-k3', reasoning: { enabled: true } },
-      });
+      expect(result).toEqual({ kind: 'ok', resolved: primaryDefaultFallback });
     }
   );
 
-  it('falls back to BALANCED_FALLBACK_MODEL when the worker returns a virtual auto model', async () => {
+  it('falls back to PRIMARY_DEFAULT_MODEL when the worker returns a virtual auto model', async () => {
     const result = await resolveAutoModel(
       {
         ...baseParams,
@@ -165,7 +163,7 @@ describe('resolveAutoModel — kilo-auto/efficient branch', () => {
       zeroBalancePromise
     );
 
-    expect(result).toEqual({ kind: 'ok', resolved: BALANCED_FALLBACK_MODEL });
+    expect(result).toEqual({ kind: 'ok', resolved: primaryDefaultFallback });
   });
 
   it('does not call the thunk more than once', async () => {
@@ -259,7 +257,7 @@ describe('resolveAutoModel — kilo-auto/efficient branch', () => {
     });
   });
 
-  it('falls back to BALANCED_FALLBACK_MODEL when variant is absent from the model catalog', async () => {
+  it('falls back to PRIMARY_DEFAULT_MODEL when variant is absent from the model catalog', async () => {
     // Claude has no "thinking" key — only none/low/medium/high/xhigh/max
     const result = await resolveAutoModel(
       {
@@ -275,10 +273,10 @@ describe('resolveAutoModel — kilo-auto/efficient branch', () => {
       zeroBalancePromise
     );
 
-    expect(result).toEqual({ kind: 'ok', resolved: BALANCED_FALLBACK_MODEL });
+    expect(result).toEqual({ kind: 'ok', resolved: primaryDefaultFallback });
   });
 
-  it('falls back to BALANCED_FALLBACK_MODEL when the model exposes no variants but decision has a variant', async () => {
+  it('falls back to PRIMARY_DEFAULT_MODEL when the model exposes no variants but decision has a variant', async () => {
     const result = await resolveAutoModel(
       {
         ...baseParams,
@@ -293,7 +291,7 @@ describe('resolveAutoModel — kilo-auto/efficient branch', () => {
       zeroBalancePromise
     );
 
-    expect(result).toEqual({ kind: 'ok', resolved: BALANCED_FALLBACK_MODEL });
+    expect(result).toEqual({ kind: 'ok', resolved: primaryDefaultFallback });
   });
 
   it('applies exact thinking and instant variant settings', async () => {
@@ -403,7 +401,7 @@ describe('resolveAutoModel — kilo-auto/efficient branch', () => {
       zeroBalancePromise
     );
 
-    expect(result).toEqual({ kind: 'ok', resolved: BALANCED_FALLBACK_MODEL });
+    expect(result).toEqual({ kind: 'ok', resolved: primaryDefaultFallback });
   });
 });
 
@@ -651,7 +649,7 @@ describe('resolveAutoModel — Organization Auto branch', () => {
 
     expect(result).toEqual({
       kind: 'ok',
-      resolved: BALANCED_FALLBACK_MODEL,
+      resolved: primaryDefaultFallback,
       routingTarget: 'kilo-auto/balanced',
     });
   });

apps/web/src/lib/ai-gateway/auto-model/resolution.ts

@@ -17,7 +17,6 @@ import {
   KILO_AUTO_BALANCED_MODEL,
   KILO_AUTO_EFFICIENT_MODEL,
   modeSchema,
-  BALANCED_FALLBACK_MODEL,
   FRONTIER_MODE_TO_MODEL,
   FRONTIER_CODE_MODEL,
   type ResolvedAutoModel,
@@ -27,6 +26,7 @@ import {
   autoFreeModels,
   findKiloExclusiveModel,
   isKiloExclusiveFreeModel,
+  PRIMARY_DEFAULT_MODEL,
   selectAutoFreeCandidate,
 } from '@/lib/ai-gateway/models';
 import { getOpenRouterModelsFromDatabase } from '@/lib/ai-gateway/providers/gateway-models-cache';
@@ -317,18 +317,19 @@ export async function resolveAutoModel(
     };
   }
   if (model === KILO_AUTO_EFFICIENT_MODEL.id || model === KILO_AUTO_BALANCED_MODEL.id) {
+    const fallbackModel = { model: PRIMARY_DEFAULT_MODEL };
     const decision = params.efficientDecision ? await params.efficientDecision() : null;
     if (decision && !isVirtualAutoModelId(decision.model)) {
       const resolvedFromDecision = await resolveEfficientDecisionModel(decision);
       if (resolvedFromDecision) {
         return { kind: 'ok', resolved: resolvedFromDecision };
       }
       // Exact catalog variant missing or removed: never serve the chosen model
-      // with implicit defaults — same balanced fallback as the no-decision path.
-      return { kind: 'ok', resolved: BALANCED_FALLBACK_MODEL };
+      // with implicit defaults — use the same fallback as the no-decision path.
+      return { kind: 'ok', resolved: fallbackModel };
     }
     // Static fallback when the worker is slow or unavailable.
-    return { kind: 'ok', resolved: BALANCED_FALLBACK_MODEL };
+    return { kind: 'ok', resolved: fallbackModel };
   }
   const mode = resolveMode(modeHeader, featureHeader);
   return {

apps/web/src/lib/ai-gateway/latest-model-aliases.ts

@@ -9,6 +9,7 @@ export const GEMINI_PRO_LATEST_MODEL_ALIAS = '~google/gemini-pro-latest';
 export const GEMINI_FLASH_LATEST_MODEL_ALIAS = '~google/gemini-flash-latest';
 export const GROK_LATEST_MODEL_ALIAS = '~x-ai/grok-latest';
 export const GLM_LATEST_MODEL_ALIAS = '~z-ai/glm-latest';
+export const GLM_FLASH_LATEST_MODEL_ALIAS = '~z-ai/glm-flash-latest';
 export const DEEPSEEK_V4_FLASH_LATEST_MODEL_ALIAS = '~deepseek/deepseek-v4-flash-latest';
 
 export const LATEST_MODEL_ALIASES = [
@@ -23,5 +24,6 @@ export const LATEST_MODEL_ALIASES = [
   GEMINI_FLASH_LATEST_MODEL_ALIAS,
   GROK_LATEST_MODEL_ALIAS,
   GLM_LATEST_MODEL_ALIAS,
+  GLM_FLASH_LATEST_MODEL_ALIAS,
   DEEPSEEK_V4_FLASH_LATEST_MODEL_ALIAS,
 ] as const;

apps/web/src/lib/ai-gateway/models.ts

@@ -8,9 +8,9 @@ import {
   claude_opus_4_7_stealth_model,
   claude_sonnet_4_6_stealth_model,
   claude_opus_4_6_stealth_model,
-  CLAUDE_SONNET_CURRENT_MODEL_ID,
   CLAUDE_OPUS_CURRENT_MODEL_ID,
 } from '@/lib/ai-gateway/providers/anthropic.constants';
+import { DEEPSEEK_V4_1_FLASH_MODEL_ID } from '@/lib/ai-gateway/providers/deepseek';
 import type { KiloExclusiveModel } from '@/lib/ai-gateway/providers/kilo-exclusive-model';
 import { isMuseModel } from '@/lib/ai-gateway/providers/meta';
 import { MINIMAX_CURRENT_MODEL_ID } from '@/lib/ai-gateway/providers/minimax';
@@ -25,12 +25,12 @@ import {
   gpt_5_6_sol_discounted_model,
   gpt_6_astra_flex_model,
 } from '@/lib/ai-gateway/providers/openai-exclusive';
-import { GLM_CURRENT_MODEL_ID } from '@/lib/ai-gateway/providers/zai';
+import { GLM_FLASH_CURRENT_MODEL_ID } from '@/lib/ai-gateway/providers/zai';
 import { type ProviderId } from '@/lib/ai-gateway/providers/types';
 import type { OpenRouterReasoningConfig } from '@/lib/ai-gateway/providers/openrouter/types';
 import { getRandomNumber } from '@/lib/ai-gateway/getRandomNumber';
 
-export const PRIMARY_DEFAULT_MODEL = CLAUDE_SONNET_CURRENT_MODEL_ID;
+export const PRIMARY_DEFAULT_MODEL = GLM_FLASH_CURRENT_MODEL_ID;
 
 export type AutoFreeModel = {
   model: string;
@@ -92,14 +92,14 @@ export const preferredModels = [
 
   ...autoFreeModels.map(({ model }) => model),
 
-  CLAUDE_SONNET_CURRENT_MODEL_ID,
   CLAUDE_OPUS_CURRENT_MODEL_ID,
   GPT_CURRENT_MODEL_ID,
   ...(gpt_5_6_sol_discounted_model.status === 'public'
     ? [gpt_5_6_sol_discounted_model.public_id]
     : []),
   ...(gpt_6_astra_flex_model.status === 'public' ? [gpt_6_astra_flex_model.public_id] : []),
-  GLM_CURRENT_MODEL_ID,
+  DEEPSEEK_V4_1_FLASH_MODEL_ID,
+  GLM_FLASH_CURRENT_MODEL_ID,
   KIMI_CURRENT_MODEL_ID,
   MINIMAX_CURRENT_MODEL_ID,
 ];

apps/web/src/lib/ai-gateway/providers/anthropic.constants.ts

@@ -7,7 +7,7 @@ export const CLAUDE_SONNET_CURRENT_MODEL_ID = 'anthropic/claude-sonnet-5';
 export const CLAUDE_OPUS_CURRENT_MODEL_ID = 'anthropic/claude-opus-5';
 export const CLAUDE_OPUS_FALLBACK_MODEL_ID = 'anthropic/claude-opus-4.8';
 export const CLAUDE_HAIKU_CURRENT_MODEL_ID = 'anthropic/claude-haiku-4.5';
-export const CLAUDE_FABLE_CURRENT_MODEL_ID = 'anthropic/claude-fable-5';
+export const CLAUDE_FABLE_CURRENT_MODEL_ID = 'anthropic/claude-fable-5.1';
 export const CLAUDE_OPUS_4_8_STEALTH_MODEL_ID = 'stealth/claude-opus-4.8';
 export const CLAUDE_OPUS_STEALTH_MODEL_ID = 'stealth/claude-opus-4.7';
 export const CLAUDE_SONNET_STEALTH_MODEL_ID = 'stealth/claude-sonnet-4.6';

apps/web/src/lib/ai-gateway/providers/deepseek.ts

@@ -1,3 +1,5 @@
 export function isDeepseekModel(model: string) {
   return model.includes('deepseek');
 }
+
+export const DEEPSEEK_V4_1_FLASH_MODEL_ID = 'deepseek/deepseek-v4.1-flash';

apps/web/src/lib/ai-gateway/providers/vercel/mapModelIdToVercel.test.ts

@@ -16,7 +16,10 @@ import {
 } from '@/lib/ai-gateway/providers/openai';
 import { mapModelIdToVercel } from '@/lib/ai-gateway/providers/vercel/mapModelIdToVercel';
 import { GROK_CURRENT_VERCEL_MODEL_ID } from '@/lib/ai-gateway/providers/xai';
-import { GLM_CURRENT_VERCEL_MODEL_ID } from '@/lib/ai-gateway/providers/zai';
+import {
+  GLM_CURRENT_VERCEL_MODEL_ID,
+  GLM_FLASH_CURRENT_VERCEL_MODEL_ID,
+} from '@/lib/ai-gateway/providers/zai';
 import {
   CLAUDE_FABLE_LATEST_MODEL_ALIAS,
   CLAUDE_HAIKU_LATEST_MODEL_ALIAS,
@@ -27,6 +30,7 @@ import {
   GEMINI_PRO_LATEST_MODEL_ALIAS,
   GPT_LATEST_MODEL_ALIAS,
   GPT_MINI_LATEST_MODEL_ALIAS,
+  GLM_FLASH_LATEST_MODEL_ALIAS,
   GLM_LATEST_MODEL_ALIAS,
   GROK_LATEST_MODEL_ALIAS,
   KIMI_LATEST_MODEL_ALIAS,
@@ -47,6 +51,7 @@ describe('mapModelIdToVercel', () => {
       [GEMINI_FLASH_LATEST_MODEL_ALIAS, GEMINI_FLASH_CURRENT_VERCEL_MODEL_ID],
       [GROK_LATEST_MODEL_ALIAS, GROK_CURRENT_VERCEL_MODEL_ID],
       [GLM_LATEST_MODEL_ALIAS, GLM_CURRENT_VERCEL_MODEL_ID],
+      [GLM_FLASH_LATEST_MODEL_ALIAS, GLM_FLASH_CURRENT_VERCEL_MODEL_ID],
       [DEEPSEEK_V4_FLASH_LATEST_MODEL_ALIAS, 'deepseek/deepseek-v4-flash-0731'],
     ])('maps %s to the current Vercel model id', (input, expected) => {
       expect(mapModelIdToVercel(input)).toBe(expected);
@@ -65,6 +70,7 @@ describe('mapModelIdToVercel', () => {
         GEMINI_FLASH_LATEST_MODEL_ALIAS,
         GROK_LATEST_MODEL_ALIAS,
         GLM_LATEST_MODEL_ALIAS,
+        GLM_FLASH_LATEST_MODEL_ALIAS,
         DEEPSEEK_V4_FLASH_LATEST_MODEL_ALIAS,
       ]);
     });

apps/web/src/lib/ai-gateway/providers/vercel/mapModelIdToVercel.ts

@@ -16,7 +16,10 @@ import {
 } from '@/lib/ai-gateway/providers/openai';
 import { inferVercelFirstPartyInferenceProviderForModel } from '@/lib/ai-gateway/providers/openrouter/inference-provider-id';
 import { GROK_CURRENT_VERCEL_MODEL_ID } from '@/lib/ai-gateway/providers/xai';
-import { GLM_CURRENT_VERCEL_MODEL_ID } from '@/lib/ai-gateway/providers/zai';
+import {
+  GLM_CURRENT_VERCEL_MODEL_ID,
+  GLM_FLASH_CURRENT_VERCEL_MODEL_ID,
+} from '@/lib/ai-gateway/providers/zai';
 import {
   CLAUDE_FABLE_LATEST_MODEL_ALIAS,
   CLAUDE_HAIKU_LATEST_MODEL_ALIAS,
@@ -27,6 +30,7 @@ import {
   GEMINI_PRO_LATEST_MODEL_ALIAS,
   GPT_LATEST_MODEL_ALIAS,
   GPT_MINI_LATEST_MODEL_ALIAS,
+  GLM_FLASH_LATEST_MODEL_ALIAS,
   GLM_LATEST_MODEL_ALIAS,
   GROK_LATEST_MODEL_ALIAS,
   KIMI_LATEST_MODEL_ALIAS,
@@ -44,6 +48,7 @@ const vercelModelIdMapping: Record<string, string | undefined> = {
   [GEMINI_FLASH_LATEST_MODEL_ALIAS]: GEMINI_FLASH_CURRENT_VERCEL_MODEL_ID,
   [GROK_LATEST_MODEL_ALIAS]: GROK_CURRENT_VERCEL_MODEL_ID,
   [GLM_LATEST_MODEL_ALIAS]: GLM_CURRENT_VERCEL_MODEL_ID,
+  [GLM_FLASH_LATEST_MODEL_ALIAS]: GLM_FLASH_CURRENT_VERCEL_MODEL_ID,
   [DEEPSEEK_V4_FLASH_LATEST_MODEL_ALIAS]: 'deepseek/deepseek-v4-flash-0731',
   'mistralai/codestral-2508': 'mistral/codestral',
   'mistralai/devstral-2512': 'mistral/devstral-2',

apps/web/src/lib/ai-gateway/providers/zai.ts

@@ -4,3 +4,5 @@ export function isGlmModel(model: string) {
 
 export const GLM_CURRENT_MODEL_ID = 'z-ai/glm-5.3';
 export const GLM_CURRENT_VERCEL_MODEL_ID = 'zai/glm-5.3';
+export const GLM_FLASH_CURRENT_MODEL_ID = 'z-ai/glm-5.3-flash';
+export const GLM_FLASH_CURRENT_VERCEL_MODEL_ID = 'zai/glm-5.3-flash';