feat(ai-gateway): add discounted GPT-5.6 Sol model

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

@@ -19,7 +19,10 @@ import {
   claude_opus_4_6_stealth_model,
 } from './providers/anthropic.constants';
 import { deepseek_v4_pro_discounted_model } from './providers/deepseek';
-import { gpt_5_6_sol_stealth_model } from './providers/openai-exclusive';
+import {
+  gpt_5_6_sol_discounted_model,
+  gpt_5_6_sol_stealth_model,
+} from './providers/openai-exclusive';
 import { tencent_hy3_free_model } from './providers/tencent';
 import { gemma_4_26b_a4b_it_free_model } from './providers/google';
 import { longcat_2_free_model } from './providers/longcat';
@@ -115,33 +118,46 @@ describe('isFreeModel', () => {
       expect(claude_opus_4_6_stealth_model.public_id).toBe('stealth/claude-opus-4.6');
     });
 
-    test('registers GPT-5.6 Sol as a Martian stealth model', () => {
-      expect(findKiloExclusiveModel('stealth/gpt-5.6-sol')).toBe(gpt_5_6_sol_stealth_model);
-      expect(gpt_5_6_sol_stealth_model.internal_id).toBe('openai/gpt-5.6-sol:optimized');
-      expect(gpt_5_6_sol_stealth_model.gateway).toBe('martian');
-      expect(getInferenceProvider(gpt_5_6_sol_stealth_model)?.slug).toBe('stealth');
-      expect(gpt_5_6_sol_stealth_model.pricing?.tiers).toEqual([
+    test('registers the discounted GPT-5.6 Sol OpenAI endpoint', async () => {
+      expect(findKiloExclusiveModel(gpt_5_6_sol_discounted_model.public_id)).toBe(
+        gpt_5_6_sol_discounted_model
+      );
+      expect(gpt_5_6_sol_discounted_model).toMatchObject({
+        internal_id: 'openai/gpt-5.6-sol',
+        gateway: 'vercel',
+        flags: ['reasoning', 'vision'],
+        inference_provider_restriction: ['openai'],
+      });
+      expect(
+        await hasBestEffortGuessDataCollectionRequirement(gpt_5_6_sol_discounted_model.public_id)
+      ).toBe(false);
+      expect(gpt_5_6_sol_discounted_model.pricing?.tiers).toEqual([
         {
           start_context_length: 0,
           pricing: {
-            prompt_per_million: 4,
-            completion_per_million: 24,
-            input_cache_read_per_million: 0.4,
-            input_cache_write_per_million: 5,
+            prompt_per_million: 2.5,
+            completion_per_million: 15,
+            input_cache_read_per_million: 0.25,
+            input_cache_write_per_million: 3.125,
           },
         },
         {
           start_context_length: 272_000,
           pricing: {
-            prompt_per_million: 8,
-            completion_per_million: 36,
-            input_cache_read_per_million: 0.8,
-            input_cache_write_per_million: 10,
+            prompt_per_million: 5,
+            completion_per_million: 22.5,
+            input_cache_read_per_million: 0.5,
+            input_cache_write_per_million: 6.25,
           },
         },
       ]);
     });
 
+    test('keeps the previous GPT-5.6 Sol stealth discount disabled', () => {
+      expect(gpt_5_6_sol_stealth_model.status).toBe('disabled');
+      expect(findKiloExclusiveModel(gpt_5_6_sol_stealth_model.public_id)).toBeNull();
+    });
+
     test('all Kilo exclusive models should have either no pricing or valid ordered pricing tiers', () => {
       for (const model of kiloExclusiveModels) {
         if (model.pricing) {
@@ -364,15 +380,18 @@ describe('getKiloExclusiveInferenceProviderRestriction', () => {
     expect(
       getKiloExclusiveInferenceProviderRestriction(deepseek_v4_pro_discounted_model.public_id)
     ).toEqual(new Set(['deepseek']));
+    expect(
+      getKiloExclusiveInferenceProviderRestriction(gpt_5_6_sol_discounted_model.public_id)
+    ).toEqual(new Set(['openai']));
     expect(getKiloExclusiveInferenceProviderRestriction(tencent_hy3_free_model.public_id)).toEqual(
       new Set(['tencent'])
     );
   });
 
   test('does not treat unrestricted exclusives or unknown ids as restricted', () => {
-    expect(
-      getKiloExclusiveInferenceProviderRestriction(gpt_5_6_sol_stealth_model.public_id)
-    ).toBeUndefined();
+    expect(getKiloExclusiveInferenceProviderRestriction(gpt_5_6_sol_stealth_model.public_id)).toBe(
+      undefined
+    );
     expect(
       getKiloExclusiveInferenceProviderRestriction(gemma_4_26b_a4b_it_free_model.public_id)
     ).toBeUndefined();

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

@@ -28,7 +28,10 @@ import { longcat_2_free_model } from '@/lib/ai-gateway/providers/longcat';
 import { isGrokModel } from '@/lib/ai-gateway/providers/xai';
 import { isClaudeModel } from '@/lib/ai-gateway/providers/anthropic.constants';
 import { GPT_CURRENT_MODEL_ID, isOpenAiModel } from '@/lib/ai-gateway/providers/openai';
-import { gpt_5_6_sol_stealth_model } from '@/lib/ai-gateway/providers/openai-exclusive';
+import {
+  gpt_5_6_sol_discounted_model,
+  gpt_5_6_sol_stealth_model,
+} from '@/lib/ai-gateway/providers/openai-exclusive';
 import { GLM_CURRENT_MODEL_ID } from '@/lib/ai-gateway/providers/zai';
 import { deepseekDiscountedModels } from '@/lib/ai-gateway/providers/deepseek';
 import { type ProviderId } from '@/lib/ai-gateway/providers/types';
@@ -97,7 +100,9 @@ export const preferredModels = [
   CLAUDE_SONNET_CURRENT_MODEL_ID,
   CLAUDE_OPUS_CURRENT_MODEL_ID,
   GPT_CURRENT_MODEL_ID,
-  ...(gpt_5_6_sol_stealth_model.status === 'public' ? [gpt_5_6_sol_stealth_model.public_id] : []),
+  ...(gpt_5_6_sol_discounted_model.status === 'public'
+    ? [gpt_5_6_sol_discounted_model.public_id]
+    : []),
   GLM_CURRENT_MODEL_ID,
   KIMI_CURRENT_MODEL_ID,
   MINIMAX_CURRENT_MODEL_ID,
@@ -134,6 +139,7 @@ export const kiloExclusiveModels = [
   ...deepseekDiscountedModels,
   qwen36_plus_stealth_model,
   gpt_5_6_sol_stealth_model,
+  gpt_5_6_sol_discounted_model,
   claude_opus_4_8_stealth_model,
   claude_opus_4_7_stealth_model,
   claude_sonnet_4_6_stealth_model,

apps/web/src/lib/ai-gateway/providers/openai-exclusive.ts

@@ -6,7 +6,7 @@ export const gpt_5_6_sol_stealth_model: KiloExclusiveModel = {
   display_name: 'Stealth: GPT-5.6 Sol (20% off)',
   description:
     "Your prompts and completions may be retained and used to train or improve the provider's services. This third-party-served variant of GPT-5.6 Sol is offered at 20% lower cost than standard GPT-5.6 Sol pricing and is not served by OpenAI or Kilo Code.",
-  status: 'public',
+  status: 'disabled',
   context_length: 1_050_000,
   max_completion_tokens: 128_000,
   gateway: 'martian',
@@ -35,3 +35,39 @@ export const gpt_5_6_sol_stealth_model: KiloExclusiveModel = {
   },
   inference_provider_restriction: [],
 };
+
+export const gpt_5_6_sol_discounted_model: KiloExclusiveModel = {
+  public_id: 'openai/gpt-5.6-sol-discounted',
+  internal_id: 'openai/gpt-5.6-sol',
+  display_name: 'OpenAI: GPT-5.6 Sol (50% off)',
+  description:
+    'GPT-5.6 Sol served by OpenAI through Vercel AI Gateway at 50% lower cost than other available inference providers. This promotion runs through September 18, 2026.',
+  status: 'public',
+  context_length: 1_050_000,
+  max_completion_tokens: 128_000,
+  gateway: 'vercel',
+  flags: ['reasoning', 'vision'],
+  pricing: {
+    tiers: [
+      {
+        start_context_length: 0,
+        pricing: {
+          prompt_per_million: 2.5,
+          completion_per_million: 15,
+          input_cache_read_per_million: 0.25,
+          input_cache_write_per_million: 3.125,
+        },
+      },
+      {
+        start_context_length: 272_000,
+        pricing: {
+          prompt_per_million: 5,
+          completion_per_million: 22.5,
+          input_cache_read_per_million: 0.5,
+          input_cache_write_per_million: 6.25,
+        },
+      },
+    ],
+  },
+  inference_provider_restriction: ['openai'],
+};

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

@@ -294,6 +294,17 @@ describe('applyVercelSettings managed requests', () => {
     expect(getVercelInferenceProvidersMock).toHaveBeenCalledWith('openai/gpt-5');
   });
 
+  it('does not add managed OpenAI BYOK to the discounted GPT-5.6 Sol endpoint', async () => {
+    process.env.OPENAI_API_KEY = 'openai-managed-key';
+    const request = managedRequest({ only: ['openai'] });
+
+    await applyManagedVercelSettings('openai/gpt-5.6-sol-discounted', request);
+
+    expect(request.body.model).toBe('openai/gpt-5.6-sol');
+    expect(request.body.providerOptions?.gateway?.only).toEqual(['openai']);
+    expect(request.body.providerOptions?.gateway?.byok).toBeUndefined();
+  });
+
   it('does not add managed OpenAI credentials when Vercel does not offer OpenAI', async () => {
     process.env.OPENAI_API_KEY = 'openai-managed-key';
     const request = managedRequest();

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

@@ -21,6 +21,7 @@ import type { AnthropicProviderOptions } from '@ai-sdk/anthropic';
 import { getRuntimeGatewayRoutingConfig } from '@/lib/ai-gateway/providers/routing-config';
 import { passesRoutingPercentage } from '@/lib/ai-gateway/providers/routing-percentage';
 import { getEnvVariable } from '@/lib/dotenvx';
+import { gpt_5_6_sol_discounted_model } from '@/lib/ai-gateway/providers/openai-exclusive';
 
 export function hasCompatibleVercelInferenceProvider(
   openRouterInferenceProviders: string[],
@@ -266,9 +267,11 @@ export async function applyVercelSettings(
 
     const gatewayOptions = requestToMutate.body.providerOptions.gateway;
     const openAiApiKey = getEnvVariable('OPENAI_API_KEY');
+    // OpenAI BYOK must also be disabled in the Vercel GUI so this model uses Vercel's discounted endpoint.
     if (
       gatewayOptions &&
       openAiApiKey &&
+      requestedModel !== gpt_5_6_sol_discounted_model.public_id &&
       vercelInferenceProviders?.includes('openai') &&
       (!gatewayOptions.only || gatewayOptions.only.includes('openai'))
     ) {

apps/web/src/tests/openrouter-models-config.test.ts

@@ -6,7 +6,7 @@ import {
 } from '@/lib/ai-gateway/providers/anthropic.constants';
 import { deepseek_v4_pro_discounted_model } from '@/lib/ai-gateway/providers/deepseek';
 import { GPT_CURRENT_MODEL_ID } from '@/lib/ai-gateway/providers/openai';
-import { gpt_5_6_sol_stealth_model } from '@/lib/ai-gateway/providers/openai-exclusive';
+import { gpt_5_6_sol_discounted_model } from '@/lib/ai-gateway/providers/openai-exclusive';
 import { QWEN37_PLUS_MODEL_ID } from '@/lib/ai-gateway/providers/qwen';
 import { tencent_hy3_free_model } from '@/lib/ai-gateway/providers/tencent';
 
@@ -37,13 +37,13 @@ describe('OpenRouter Models Config', () => {
       expect(preferredModels).not.toContain(model);
     });
 
-    if (gpt_5_6_sol_stealth_model.status === 'public') {
-      expect(preferredModels).toContain(gpt_5_6_sol_stealth_model.public_id);
+    if (gpt_5_6_sol_discounted_model.status === 'public') {
+      expect(preferredModels).toContain(gpt_5_6_sol_discounted_model.public_id);
       expect(preferredModels.indexOf(GPT_CURRENT_MODEL_ID)).toBeLessThan(
-        preferredModels.indexOf(gpt_5_6_sol_stealth_model.public_id)
+        preferredModels.indexOf(gpt_5_6_sol_discounted_model.public_id)
       );
     } else {
-      expect(preferredModels).not.toContain(gpt_5_6_sol_stealth_model.public_id);
+      expect(preferredModels).not.toContain(gpt_5_6_sol_discounted_model.public_id);
     }
   });
 });