Add GPT-6 Astra Flex model

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

@@ -18,7 +18,7 @@ import {
   claude_sonnet_4_6_stealth_model,
   claude_opus_4_6_stealth_model,
 } from './providers/anthropic.constants';
-import { gpt_5_6_sol_discounted_model } from './providers/openai-exclusive';
+import { gpt_5_6_sol_discounted_model, gpt_6_astra_flex_model } from './providers/openai-exclusive';
 import { gemma_4_26b_a4b_it_free_model } from './providers/google';
 import { isUnavailableModel } from './unavailable-models';
 import { getRandomNumber } from './getRandomNumber';
@@ -152,6 +152,41 @@ describe('isFreeModel', () => {
       ]);
     });
 
+    test('keeps the GPT-6 Astra OpenAI Flex endpoint disabled', async () => {
+      expect(findKiloExclusiveModel(gpt_6_astra_flex_model.public_id)).toBeNull();
+      expect(gpt_6_astra_flex_model).toMatchObject({
+        status: 'disabled',
+        internal_id: 'openai/gpt-6-astra',
+        gateway: 'vercel',
+        flags: ['reasoning', 'vision', 'flex'],
+        inference_provider_restriction: ['openai'],
+        pricing: { fallbackOnly: true },
+      });
+      expect(
+        await hasBestEffortGuessDataCollectionRequirement(gpt_6_astra_flex_model.public_id)
+      ).toBe(false);
+      expect(gpt_6_astra_flex_model.pricing?.tiers).toEqual([
+        {
+          start_context_length: 0,
+          pricing: {
+            prompt_per_million: 5,
+            completion_per_million: 25,
+            input_cache_read_per_million: 0.5,
+            input_cache_write_per_million: 6.25,
+          },
+        },
+        {
+          start_context_length: 272_000,
+          pricing: {
+            prompt_per_million: 10,
+            completion_per_million: 37.5,
+            input_cache_read_per_million: 1,
+            input_cache_write_per_million: 12.5,
+          },
+        },
+      ]);
+    });
+
     test('all Kilo exclusive models should have either no pricing or valid ordered pricing tiers', () => {
       for (const model of kiloExclusiveModels) {
         if (model.pricing) {

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

@@ -21,7 +21,10 @@ import { stepfun_37_flash_free_model } from '@/lib/ai-gateway/providers/stepfun'
 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_discounted_model } from '@/lib/ai-gateway/providers/openai-exclusive';
+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 { type ProviderId } from '@/lib/ai-gateway/providers/types';
 import type { OpenRouterReasoningConfig } from '@/lib/ai-gateway/providers/openrouter/types';
@@ -95,6 +98,7 @@ export const preferredModels = [
   ...(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,
   KIMI_CURRENT_MODEL_ID,
   MINIMAX_CURRENT_MODEL_ID,
@@ -130,6 +134,7 @@ export const kiloExclusiveModels = [
   gemma_4_26b_a4b_it_free_model,
   qwen36_plus_stealth_model,
   gpt_5_6_sol_discounted_model,
+  gpt_6_astra_flex_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/apply-provider-specific-logic.test.ts

@@ -1,10 +1,11 @@
-import { describe, expect, it } from '@jest/globals';
+import { describe, expect, it, jest } from '@jest/globals';
 import { CLAUDE_OPUS_FALLBACK_MODEL_ID } from '@/lib/ai-gateway/providers/anthropic.constants';
 import {
   applyAnthropicThinkingDefault,
   applyGatewayModelsFallback,
   applyPreferredProvider,
   applyReasoningDetailsTransform,
+  removeUnsupportedRequestServiceTier,
 } from '@/lib/ai-gateway/providers/apply-provider-specific-logic';
 import type { GatewayRequest } from '@/lib/ai-gateway/providers/openrouter/types';
 import {
@@ -13,6 +14,11 @@ import {
   type ProviderId,
 } from '@/lib/ai-gateway/providers/types';
 import { PERPLEXITY_KIMI_PUBLIC_ID } from '@/lib/ai-gateway/providers/partner/constants';
+import { QWEN37_MAX_MODEL_ID } from '@/lib/ai-gateway/custom-pricing';
+import {
+  gpt_5_6_sol_discounted_model,
+  gpt_6_astra_flex_model,
+} from '@/lib/ai-gateway/providers/openai-exclusive';
 
 function makeRequest(model: string, models?: string[]): GatewayRequest {
   return {
@@ -85,6 +91,55 @@ describe('applyAnthropicThinkingDefault', () => {
   );
 });
 
+describe('removeUnsupportedRequestServiceTier', () => {
+  it.each([
+    {
+      model: QWEN37_MAX_MODEL_ID,
+      kiloExclusiveModel: null,
+      reason: 'non-fallback custom pricing',
+    },
+    {
+      model: gpt_5_6_sol_discounted_model.public_id,
+      kiloExclusiveModel: gpt_5_6_sol_discounted_model,
+      reason: 'non-Flex Kilo-exclusive model',
+    },
+  ])(
+    'removes and logs the request-level tier for $reason',
+    ({ model, kiloExclusiveModel, reason }) => {
+      const request = makeRequest(model);
+      request.body.service_tier = 'priority';
+      const warn = jest.spyOn(console, 'warn').mockImplementation(() => undefined);
+
+      removeUnsupportedRequestServiceTier(model, request, kiloExclusiveModel);
+
+      expect(request.body.service_tier).toBeUndefined();
+      expect(warn).toHaveBeenCalledWith(
+        '[applyProviderSpecificLogic] Removed unsupported request-level service tier',
+        {
+          model,
+          requestKind: 'chat_completions',
+          serviceTier: 'priority',
+          reason,
+        }
+      );
+      warn.mockRestore();
+    }
+  );
+
+  it.each([
+    [PERPLEXITY_KIMI_PUBLIC_ID, null],
+    [gpt_6_astra_flex_model.public_id, gpt_6_astra_flex_model],
+    ['vendor/standard-model', null],
+  ] as const)('preserves the request-level tier for %s', (model, kiloExclusiveModel) => {
+    const request = makeRequest(model);
+    request.body.service_tier = 'priority';
+
+    removeUnsupportedRequestServiceTier(model, request, kiloExclusiveModel);
+
+    expect(request.body.service_tier).toBe('priority');
+  });
+});
+
 describe('applyReasoningDetailsTransform', () => {
   function makeProvider(responseTransforms: Provider['responseTransforms']): Provider {
     return {

apps/web/src/lib/ai-gateway/providers/apply-provider-specific-logic.ts

@@ -7,7 +7,10 @@ import {
 } from '@/lib/ai-gateway/providers/openrouter/types';
 import { applyMistralModelSettings, isMistralModel } from '@/lib/ai-gateway/providers/mistral';
 import { findKiloExclusiveModel } from '@/lib/ai-gateway/models';
-import { applyKiloExclusiveModelSettings } from '@/lib/ai-gateway/providers/kilo-exclusive-model';
+import {
+  applyKiloExclusiveModelSettings,
+  type KiloExclusiveModel,
+} from '@/lib/ai-gateway/providers/kilo-exclusive-model';
 import { applyAnthropicModelSettings } from '@/lib/ai-gateway/providers/anthropic';
 import {
   CLAUDE_OPUS_FALLBACK_MODEL_ID,
@@ -49,6 +52,7 @@ import { isFreeModel } from '@/lib/ai-gateway/is-free-model';
 import { isOpenAiModel } from '@/lib/ai-gateway/providers/openai';
 import { ReasoningFormat } from '@/lib/ai-gateway/custom-llm/format';
 import { ReasoningDetailType } from '@/lib/ai-gateway/custom-llm/reasoning-details';
+import { getCustomPricing } from '@/lib/ai-gateway/custom-pricing';
 
 function isRecord(value: unknown): value is Record<string, unknown> {
   return typeof value === 'object' && value !== null && !Array.isArray(value);
@@ -223,6 +227,32 @@ export function applyAnthropicThinkingDefault(
   }
 }
 
+export function removeUnsupportedRequestServiceTier(
+  requestedModel: string,
+  requestToMutate: GatewayRequest,
+  kiloExclusiveModel: KiloExclusiveModel | null
+) {
+  const customPricing = getCustomPricing(requestedModel);
+  const reason =
+    customPricing && !customPricing.fallbackOnly
+      ? 'non-fallback custom pricing'
+      : kiloExclusiveModel && !kiloExclusiveModel.flags.includes('flex')
+        ? 'non-Flex Kilo-exclusive model'
+        : null;
+  const serviceTier = requestToMutate.body.service_tier;
+  if (!reason || serviceTier === undefined) {
+    return;
+  }
+
+  console.warn('[applyProviderSpecificLogic] Removed unsupported request-level service tier', {
+    model: requestedModel,
+    requestKind: requestToMutate.kind,
+    serviceTier,
+    reason,
+  });
+  delete requestToMutate.body.service_tier;
+}
+
 /**
  * Inverse of the reasoning-content response transform: folds
  * client-supplied `reasoning_details` back into the `reasoning_content` string
@@ -288,6 +318,7 @@ export async function applyProviderSpecificLogic(
   enableReasoningSummaries(requestToMutate);
 
   const kiloExclusiveModel = findKiloExclusiveModel(requestedModel);
+  removeUnsupportedRequestServiceTier(requestedModel, requestToMutate, kiloExclusiveModel);
   if (kiloExclusiveModel) {
     applyKiloExclusiveModelSettings(requestToMutate, kiloExclusiveModel);
   }

apps/web/src/lib/ai-gateway/providers/kilo-exclusive-model.test.ts

@@ -8,7 +8,9 @@ import {
   type PricingTiers,
 } from '@/lib/ai-gateway/providers/kilo-exclusive-model';
 import type {
+  GatewayMessagesRequest,
   GatewayRequest,
+  GatewayResponsesRequest,
   OpenRouterChatCompletionRequest,
   OpenRouterProviderConfig,
 } from '@/lib/ai-gateway/providers/openrouter/types';
@@ -157,6 +159,28 @@ describe('applyKiloExclusiveModelSettings', () => {
     expect(req.body.model).toBe('vendor/real-model');
   });
 
+  it.each([
+    {
+      kind: 'chat_completions' as const,
+      body: { model: 'kilo/test-model', messages: [] } as OpenRouterChatCompletionRequest,
+    },
+    {
+      kind: 'responses' as const,
+      body: { model: 'kilo/test-model', input: '' } as GatewayResponsesRequest,
+    },
+    {
+      kind: 'messages' as const,
+      body: { model: 'kilo/test-model', max_tokens: 1, messages: [] } as GatewayMessagesRequest,
+    },
+  ])('sets the Flex service tier for $kind requests', request => {
+    applyKiloExclusiveModelSettings(
+      request,
+      makeModel({ internal_id: 'vendor/real-model', flags: ['flex'] })
+    );
+
+    expect(request.body.service_tier).toBe('flex');
+  });
+
   it('leaves provider untouched when there is no restriction', () => {
     const req = makeRequest({ only: ['anthropic'], zdr: true });
     applyKiloExclusiveModelSettings(req, makeModel({ internal_id: 'vendor/x' }));

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

@@ -11,6 +11,7 @@ import {
 export type KiloExclusiveModelFlag =
   | 'reasoning'
   | 'vision'
+  | 'flex'
   | 'stealth'
   | 'vercel-routing'
   | 'rate-limited'
@@ -169,6 +170,9 @@ export function applyKiloExclusiveModelSettings(
 ) {
   requestToMutate.body.model = kiloExclusiveModel.internal_id;
   removeNonSensicalMaxTokens(requestToMutate, kiloExclusiveModel);
+  if (kiloExclusiveModel.flags.includes('flex')) {
+    requestToMutate.body.service_tier = 'flex';
+  }
   const restriction = kiloExclusiveModel.inference_provider_restriction;
   if (restriction.length === 0) {
     return;

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

@@ -35,3 +35,40 @@ export const gpt_5_6_sol_discounted_model: KiloExclusiveModel = {
   },
   inference_provider_restriction: ['openai'],
 };
+
+export const gpt_6_astra_flex_model: KiloExclusiveModel = {
+  public_id: 'openai/gpt-6-astra-flex',
+  internal_id: 'openai/gpt-6-astra',
+  display_name: 'OpenAI: GPT-6 Astra Flex',
+  description:
+    'GPT-6 Astra with OpenAI Flex processing, offering lower costs in exchange for slower response times and occasional resource unavailability.',
+  status: 'disabled',
+  context_length: 1_050_000,
+  max_completion_tokens: 128_000,
+  gateway: 'vercel',
+  flags: ['reasoning', 'vision', 'flex'],
+  pricing: {
+    fallbackOnly: true,
+    tiers: [
+      {
+        start_context_length: 0,
+        pricing: {
+          prompt_per_million: 5,
+          completion_per_million: 25,
+          input_cache_read_per_million: 0.5,
+          input_cache_write_per_million: 6.25,
+        },
+      },
+      {
+        start_context_length: 272_000,
+        pricing: {
+          prompt_per_million: 10,
+          completion_per_million: 37.5,
+          input_cache_read_per_million: 1,
+          input_cache_write_per_million: 12.5,
+        },
+      },
+    ],
+  },
+  inference_provider_restriction: ['openai'],
+};

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

@@ -178,6 +178,7 @@ export function convertProviderOptions(
     );
   })();
 
+  const serviceTier = requestToMutate.body.service_tier;
   return {
     gateway: {
       only,
@@ -186,6 +187,7 @@ export function convertProviderOptions(
       zeroDataRetention: provider?.zdr,
       disallowPromptTraining: provider?.data_collection === 'deny' || undefined,
       models: requestToMutate.body.models,
+      serviceTier: serviceTier === 'flex' || serviceTier === 'priority' ? serviceTier : undefined,
     },
   };
 }

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

@@ -12,7 +12,10 @@ import {
   CLAUDE_SONNET_CURRENT_MODEL_ID,
 } from '@/lib/ai-gateway/providers/anthropic.constants';
 import { GPT_CURRENT_MODEL_ID } from '@/lib/ai-gateway/providers/openai';
-import { gpt_5_6_sol_discounted_model } from '@/lib/ai-gateway/providers/openai-exclusive';
+import {
+  gpt_5_6_sol_discounted_model,
+  gpt_6_astra_flex_model,
+} from '@/lib/ai-gateway/providers/openai-exclusive';
 import {
   GEMMA_4_26B_A4B_IT_ID,
   gemma_4_26b_a4b_it_free_model,
@@ -59,6 +62,12 @@ describe('OpenRouter Models Config', () => {
     } else {
       expect(preferredModels).not.toContain(gpt_5_6_sol_discounted_model.public_id);
     }
+
+    if (gpt_6_astra_flex_model.status === 'public') {
+      expect(preferredModels).toContain(gpt_6_astra_flex_model.public_id);
+    } else {
+      expect(preferredModels).not.toContain(gpt_6_astra_flex_model.public_id);
+    }
   });
 
   test('monitors only concrete preferred models', () => {