feat(vercel): use -non-reasoning suffix for Grok 4.x when reasoning is disabled

apps/web/src/app/api/openrouter/embeddings/route.ts

@@ -229,7 +229,7 @@ export async function POST(request: NextRequest): Promise<NextResponseType<unkno
   const effectiveProvider = provider;
 
   if (userByok && userByok.length > 0 && provider.id === 'vercel') {
-    requestBodyParsed.model = mapModelIdToVercel(requestBodyParsed.model);
+    requestBodyParsed.model = mapModelIdToVercel(requestBodyParsed.model, false);
   }
 
   const upstreamBody = buildUpstreamBody(requestBodyParsed);

apps/web/src/lib/ai-gateway/byok/index.ts

@@ -37,7 +37,7 @@ export async function getModelUserByokProviders(modelId: string): Promise<UserBy
     return [];
   }
   const providers =
-    vercelModelMetadata[mapModelIdToVercel(modelId)]?.endpoints
+    vercelModelMetadata[mapModelIdToVercel(modelId, false)]?.endpoints
       .map(ep => VercelUserByokInferenceProviderIdSchema.safeParse(ep.tag).data)
       .filter(providerId => providerId !== undefined) ?? [];
   if (providers.length === 0) {

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

@@ -12,6 +12,7 @@ import type {
   VercelInferenceProviderConfig,
   VercelProviderConfig,
 } from '@/lib/ai-gateway/providers/openrouter/types';
+import { isReasoningExplicitlyDisabled } from '@/lib/ai-gateway/providers/openrouter/request-helpers';
 import { mapModelIdToVercel } from '@/lib/ai-gateway/providers/vercel/mapModelIdToVercel';
 import { StoredModelSchema } from '@kilocode/db';
 import * as z from 'zod';
@@ -73,7 +74,7 @@ export async function shouldRouteToVercel(
   }
 
   const vercelModels = await getVercelModels();
-  const vercelModelId = mapModelIdToVercel(requestedModel);
+  const vercelModelId = mapModelIdToVercel(requestedModel, isReasoningExplicitlyDisabled(request));
   if (!vercelModels.includes(vercelModelId)) {
     console.debug(`[shouldRouteToVercel] model not found in Vercel model list`);
     return false;
@@ -135,7 +136,10 @@ export function applyVercelSettings(
   requestToMutate: GatewayRequest,
   userByok: BYOKResult[] | null
 ) {
-  requestToMutate.body.model = mapModelIdToVercel(requestedModel);
+  requestToMutate.body.model = mapModelIdToVercel(
+    requestedModel,
+    isReasoningExplicitlyDisabled(requestToMutate)
+  );
 
   if (userByok) {
     if (userByok.length === 0) {

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

@@ -26,9 +26,12 @@ const vercelModelIdMapping: Record<string, string | undefined> = {
   'qwen/qwen3-32b': 'alibaba/qwen-3-32b',
 };
 
-export function mapModelIdToVercel(modelId: string) {
+export function mapModelIdToVercel(modelId: string, reasoningExplicitlyDisabled: boolean) {
   const hardcodedVercelId = vercelModelIdMapping[modelId];
   if (hardcodedVercelId) {
+    if (reasoningExplicitlyDisabled && hardcodedVercelId.endsWith('-reasoning')) {
+      return hardcodedVercelId.replace(/-reasoning$/, '-non-reasoning');
+    }
     return hardcodedVercelId;
   }
 

apps/web/src/routers/byok-router.ts

@@ -80,7 +80,7 @@ async function fetchSupportedModels() {
   result['codestral'] = ['Codestral (mistralai/codestral-2508)'];
 
   for (const openRouterModel of Object.values(openRouterModelMetadata.data)) {
-    const vercelModel = vercelModelMetadata.data[mapModelIdToVercel(openRouterModel.id)];
+    const vercelModel = vercelModelMetadata.data[mapModelIdToVercel(openRouterModel.id, false)];
     if (!vercelModel) continue;
     if (vercelModel.id.includes('codestral')) continue;
     if (vercelModel.type !== 'language') continue;