feat(models): onboard GPT-6 Astra, GPT-6.1 Sol, GPT-6 Sol, GPT-6 Luna

src/integrations/generated/integrationArtifacts.generated.ts

@@ -77,6 +77,7 @@ import modelFireworksMerged from '../models/fireworks-merged.js'
 import modelGemini from '../models/gemini.js'
 import modelGlm from '../models/glm.js'
 import modelGpt from '../models/gpt.js'
+import modelGpt6 from '../models/gpt6.js'
 import modelKimi from '../models/kimi.js'
 import modelLing from '../models/ling.js'
 import modelLlama from '../models/llama.js'
@@ -97,7 +98,7 @@ export const VENDOR_DESCRIPTORS = [vendorAnthropic, vendorBankr, vendorDeepseek,
 export const GATEWAY_DESCRIPTORS = [gatewayAimlapi, gatewayApiRoute, gatewayApismart, gatewayAtlasCloud, gatewayAtomicChat, gatewayAzureOpenai, gatewayBedrock, gatewayClinepass, gatewayCloudflare, gatewayCommandcode, gatewayConcentrate, gatewayCustom, gatewayDashscopeCn, gatewayDashscopeIntl, gatewayGithubEnterprise, gatewayGithub, gatewayGitlawbOpengateway, gatewayGroq, gatewayHicap, gatewayKimiCode, gatewayLlmtr, gatewayLmstudio, gatewayMistral, gatewayNvidiaNim, gatewayOllama, gatewayOpencodeGo, gatewayOpencode, gatewayOpenrouter, gatewayTogether, gatewayVertex, gatewayXiaomiMimoToken] as const satisfies readonly GatewayDescriptor[]
 export const ANTHROPIC_PROXY_DESCRIPTORS = [anthropicproxyCustom] as const satisfies readonly AnthropicProxyDescriptor[]
 export const BRAND_DESCRIPTORS = [brandClaude, brandDeepseek, brandFireworks, brandGemini, brandGlm, brandGpt, brandKimi, brandLing, brandLlama, brandLongcat, brandMacaron, brandMinimax, brandMistral, brandNearai, brandNemotron, brandOpenaiCompatibleAlias, brandQwen, brandTencent, brandXai, brandXiaomiMimo] as const satisfies readonly BrandDescriptor[]
-export const MODEL_DESCRIPTOR_GROUPS = [modelClaude, modelDeepseek, modelFireworksMerged, modelGemini, modelGlm, modelGpt, modelKimi, modelLing, modelLlama, modelLongcat, modelMacaron, modelMinimax, modelMistral, modelNearai, modelNemotron, modelOpenaiCompatibleAlias, modelOpencode, modelQwen, modelTencent, modelXai, modelXiaomiMimo] as const satisfies readonly (readonly ModelDescriptor[])[]
+export const MODEL_DESCRIPTOR_GROUPS = [modelClaude, modelDeepseek, modelFireworksMerged, modelGemini, modelGlm, modelGpt, modelGpt6, modelKimi, modelLing, modelLlama, modelLongcat, modelMacaron, modelMinimax, modelMistral, modelNearai, modelNemotron, modelOpenaiCompatibleAlias, modelOpencode, modelQwen, modelTencent, modelXai, modelXiaomiMimo] as const satisfies readonly (readonly ModelDescriptor[])[]
 export const MODEL_DESCRIPTORS = MODEL_DESCRIPTOR_GROUPS.flat() satisfies readonly ModelDescriptor[]
 
 export { PROVIDER_PRESET_MANIFEST, ORDERED_PROVIDER_PRESETS } from './integrationManifest.generated.js'

src/integrations/gpt6Catalog.test.ts

@@ -0,0 +1,43 @@
+import { expect, test } from 'bun:test'
+
+import gpt6Models from './models/gpt6.js'
+import openaiVendor from './vendors/openai.js'
+
+const GPT6_FLAGSHIPS = ['gpt-6-astra', 'gpt-6.1-sol', 'gpt-6-sol'] as const
+const GPT6_IDS = [...GPT6_FLAGSHIPS, 'gpt-6-luna'] as const
+
+test('gpt-6 model descriptors carry the full API spec (no Codex cap published)', () => {
+  for (const id of GPT6_IDS) {
+    const descriptor = gpt6Models.find(model => model.id === id)
+    expect(descriptor).toBeDefined()
+    // Descriptor and catalog intentionally AGREE here — unlike the 5.6
+    // family, there is no published Codex-route cap for GPT-6 (the 272k pin
+    // is verified ONLY for 5.6; see the gpt6.ts descriptor comment). The
+    // full 1.05M spec ships with a dated comment + first-500 monitoring
+    // flag instead of an invented cap.
+    expect(descriptor?.contextWindow).toBe(1_050_000)
+    expect(descriptor?.maxOutputTokens).toBe(128_000)
+  }
+})
+
+test('openai vendor gpt-6 catalog entries carry limits and reasoning metadata', () => {
+  for (const id of GPT6_IDS) {
+    const entry = openaiVendor.catalog?.models?.find(model => model.id === id)
+    expect(entry).toBeDefined()
+    expect(entry?.contextWindow).toBe(1_050_000)
+    expect(entry?.maxOutputTokens).toBe(128_000)
+    expect(entry?.reasoning?.wireFormat).toBe('reasoning_effort')
+    expect(entry?.reasoning?.levels).toContain('xhigh')
+  }
+})
+
+test('openai vendor gpt-6 flagships expose the Ultra tier as max; luna tops out at xhigh', () => {
+  for (const id of GPT6_FLAGSHIPS) {
+    const entry = openaiVendor.catalog?.models?.find(model => model.id === id)
+    expect(entry?.reasoning?.levels).toContain('max')
+    expect(entry?.reasoning?.defaultLevel).toBe('high')
+  }
+  const luna = openaiVendor.catalog?.models?.find(model => model.id === 'gpt-6-luna')
+  expect(luna?.reasoning?.levels).not.toContain('max')
+  expect(luna?.reasoning?.defaultLevel).toBe('medium')
+})

src/integrations/models/gpt6.ts

@@ -0,0 +1,56 @@
+import { defineModel } from '../define.js'
+
+const gpt6Capabilities = {
+  supportsVision: true,
+  supportsStreaming: true,
+  supportsFunctionCalling: true,
+  supportsJsonMode: true,
+  supportsReasoning: false,
+  supportsPreciseTokenCount: true,
+}
+
+function gpt6Model(
+  id: string,
+  label: string,
+  contextWindow: number,
+  maxOutputTokens: number,
+) {
+  return defineModel({
+    id,
+    label,
+    brandId: 'gpt',
+    vendorId: 'openai',
+    classification: ['chat', 'vision', 'coding'],
+    defaultModel: id,
+    capabilities: gpt6Capabilities,
+    contextWindow,
+    maxOutputTokens,
+  })
+}
+
+export default [
+  // GPT-6 family (October 2026). Live API spec, two agreeing sources
+  // (2026-10-13): OpenRouter (`openai/gpt-6-astra`: `context_length`
+  // 1,050,000, `top_provider.max_completion_tokens` 128,000) and models.dev
+  // (`limit.context` 1,050,000, `limit.input` 922,000, `limit.output`
+  // 128,000). The same 1,050,000 / 128,000 spec is verified for gpt-6.1-sol,
+  // gpt-6-sol, and gpt-6-luna (both sources).
+  //
+  // Unlike the 5.6 family, there is NO published Codex-route cap for these
+  // models: OpenAI's Codex bundled metadata pin (openai/codex#33972) covers
+  // the 5.6 family only, and no Astra/Sol/Luna Codex number (or pricing-tier
+  // boundary rule) could be found as of 2026-10-13 — in-session research came
+  // back empty (no web tooling available; zero Astra references in-tree).
+  // So the descriptors below carry the full API spec instead of a
+  // Codex-route pin. Do NOT copy the 5.6 272k cap here: that number is
+  // verified ONLY for the 5.6 family. FIRST-500 MONITORING: watch early
+  // production Codex traffic for 500 "input exceeds the context window"
+  // failures (issue #1118 precedent); if the Codex backend enforces a lower
+  // effective cap, pin these descriptors to it with provenance.
+  // The 1.05M API spec lives in the openai-vendor catalog entries, same as
+  // the 5.6 family. Last cross-checked 2026-10-13.
+  gpt6Model('gpt-6-astra', 'GPT-6 Astra', 1_050_000, 128_000),
+  gpt6Model('gpt-6.1-sol', 'GPT-6.1 Sol', 1_050_000, 128_000),
+  gpt6Model('gpt-6-sol', 'GPT-6 Sol', 1_050_000, 128_000),
+  gpt6Model('gpt-6-luna', 'GPT-6 Luna', 1_050_000, 128_000),
+]

src/integrations/vendors/openai.ts

@@ -63,6 +63,21 @@ export default defineVendor({
       { id: 'gpt-5.6-terra', apiName: 'gpt-5.6-terra', label: 'GPT-5.6 Terra', modelDescriptorId: 'gpt-5.6-terra', contextWindow: 1_050_000, maxOutputTokens: 128_000, capabilities: { supportsReasoning: true }, reasoning: { mode: 'levels', levels: ['low', 'medium', 'high', 'xhigh'], defaultLevel: 'high', wireFormat: 'reasoning_effort' } },
       { id: 'gpt-5.6-luna', apiName: 'gpt-5.6-luna', label: 'GPT-5.6 Luna', modelDescriptorId: 'gpt-5.6-luna', contextWindow: 1_050_000, maxOutputTokens: 128_000, capabilities: { supportsReasoning: true }, reasoning: { mode: 'levels', levels: ['low', 'medium', 'high', 'xhigh'], defaultLevel: 'high', wireFormat: 'reasoning_effort' } },
       { id: 'gpt-5.5', apiName: 'gpt-5.5', label: 'GPT-5.5', modelDescriptorId: 'gpt-5.5', contextWindow: 272_000, maxOutputTokens: 128_000, capabilities: { supportsReasoning: true }, reasoning: { mode: 'levels', levels: ['low', 'medium', 'high', 'xhigh'], defaultLevel: 'high', wireFormat: 'reasoning_effort' } },
+      // GPT-6 family (October 2026): same /v1/responses + nested
+      // reasoning.effort wire shape as the 5.6 family (modelRequiresResponsesApi
+      // routes them there; verified 5.6 behavior, inherited pending live
+      // GPT-6 wire confirmation). Effort ladder per models.dev
+      // `reasoning_options` (low, medium, high, xhigh, max) plus the
+      // operator-confirmed native-Codex Ultra tier on Sol/Astra model types
+      // only (2026-10-13): flagships (astra/sol/6.1-sol) expose `max`, Luna
+      // tops out at `xhigh`. The Ultra→max wire mapping is the working
+      // hypothesis, not a live-verified row — see the gpt6.ts descriptor
+      // comment. Defaults inherit the family pattern (high flagship,
+      // medium luna) pending Codex-default evidence.
+      { id: 'gpt-6-astra', apiName: 'gpt-6-astra', label: 'GPT-6 Astra', modelDescriptorId: 'gpt-6-astra', contextWindow: 1_050_000, maxOutputTokens: 128_000, capabilities: { supportsReasoning: true }, reasoning: { mode: 'levels', levels: ['low', 'medium', 'high', 'xhigh', 'max'], defaultLevel: 'high', wireFormat: 'reasoning_effort' } },
+      { id: 'gpt-6.1-sol', apiName: 'gpt-6.1-sol', label: 'GPT-6.1 Sol', modelDescriptorId: 'gpt-6.1-sol', contextWindow: 1_050_000, maxOutputTokens: 128_000, capabilities: { supportsReasoning: true }, reasoning: { mode: 'levels', levels: ['low', 'medium', 'high', 'xhigh', 'max'], defaultLevel: 'high', wireFormat: 'reasoning_effort' } },
+      { id: 'gpt-6-sol', apiName: 'gpt-6-sol', label: 'GPT-6 Sol', modelDescriptorId: 'gpt-6-sol', contextWindow: 1_050_000, maxOutputTokens: 128_000, capabilities: { supportsReasoning: true }, reasoning: { mode: 'levels', levels: ['low', 'medium', 'high', 'xhigh', 'max'], defaultLevel: 'high', wireFormat: 'reasoning_effort' } },
+      { id: 'gpt-6-luna', apiName: 'gpt-6-luna', label: 'GPT-6 Luna', modelDescriptorId: 'gpt-6-luna', contextWindow: 1_050_000, maxOutputTokens: 128_000, capabilities: { supportsReasoning: true }, reasoning: { mode: 'levels', levels: ['low', 'medium', 'high', 'xhigh'], defaultLevel: 'medium', wireFormat: 'reasoning_effort' } },
       { id: 'gpt-5.4', apiName: 'gpt-5.4', label: 'GPT-5.4', modelDescriptorId: 'gpt-5.4', contextWindow: 1_050_000, maxOutputTokens: 128_000, capabilities: { supportsReasoning: true }, reasoning: { mode: 'levels', levels: ['low', 'medium', 'high', 'xhigh'], defaultLevel: 'high', wireFormat: 'reasoning_effort' } },
       { id: 'gpt-5-mini', apiName: 'gpt-5-mini', label: 'GPT-5 Mini' },
       { id: 'gpt-4o', apiName: 'gpt-4o', label: 'GPT-4o' },

src/services/api/providerConfig.local.test.ts

@@ -380,6 +380,13 @@ test('modelRequiresResponsesApi matches gpt-5.4/5.5/5.6 (excl. mini/nano) only',
     'gpt-5.6-terra',
     'gpt-5.6-luna',
     'GPT-5.6-SOL',
+    // GPT-6 family inherits the 5.6 responses routing
+    'gpt-6',
+    'gpt-6-astra',
+    'gpt-6.1-sol',
+    'gpt-6-sol',
+    'gpt-6-luna',
+    'GPT-6-ASTRA',
     // patch releases of a verified family stay routed
     'gpt-5.4.1',
   ]) {
@@ -402,6 +409,9 @@ test('modelRequiresResponsesApi matches gpt-5.4/5.5/5.6 (excl. mini/nano) only',
     'gpt-5.7',
     'gpt-5.8-preview',
     'gpt-5.9',
+    // gpt-60-style ids are not the gpt-6 family
+    'gpt-60',
+    'gpt-6-mini',
     'o3',
     'claude-opus-4-8',
   ]) {
@@ -434,6 +444,20 @@ test('auto-routes gpt-5.6 to responses on the default OpenAI base', () => {
   })
 })
 
+test('auto-routes gpt-6-astra to responses on the default OpenAI base', () => {
+  process.env.CLAUDE_CODE_USE_OPENAI = '1'
+  process.env.OPENAI_BASE_URL = 'https://api.openai.com/v1'
+  process.env.OPENAI_MODEL = 'gpt-6-astra'
+  delete process.env.OPENAI_API_FORMAT
+
+  expect(resolveProviderRequest()).toMatchObject({
+    transport: 'responses',
+    resolvedModel: 'gpt-6-astra',
+    baseUrl: 'https://api.openai.com/v1',
+    reasoning: { effort: 'high' },
+  })
+})
+
 test('auto-routes gpt-5.6 to responses on regional OpenAI subdomains', () => {
   process.env.CLAUDE_CODE_USE_OPENAI = '1'
   process.env.OPENAI_BASE_URL = 'https://eu.api.openai.com/v1'

src/services/api/providerConfig.protoAlias.test.ts

@@ -71,6 +71,16 @@ describe('providerConfig — Codex alias lookup is prototype-safe', () => {
     expect(getReasoningEffortForModel('gpt-5.3-codex-spark')).toBeUndefined()
   })
 
+  test('getReasoningEffortForModel resolves GPT-6 family defaults', () => {
+    expect(getReasoningEffortForModel('gpt-6')).toBe('high')
+    expect(getReasoningEffortForModel('gpt-6-astra')).toBe('high')
+    expect(getReasoningEffortForModel('gpt-6.1-sol')).toBe('high')
+    expect(getReasoningEffortForModel('gpt-6-sol')).toBe('high')
+    expect(getReasoningEffortForModel('gpt-6-luna')).toBe('medium')
+    expect(supportsCodexReasoningEffort('gpt-6-astra')).toBe(true)
+    expect(supportsCodexReasoningEffort('gpt-6-luna')).toBe(true)
+  })
+
   // The descriptor parsing path (parseModelDescriptor, via resolveProviderRequest)
   // is guarded too — a proto-name model must resolve to itself, not to the
   // inherited Object constructor's (nonexistent) `.model`, which produced an

src/services/api/providerConfig.ts

@@ -103,6 +103,33 @@ const CODEX_ALIAS_MODELS: Record<
     model: 'gpt-5.6-luna',
     reasoningEffort: 'medium',
   },
+  // GPT-6 family (October 2026). Served on the Codex backend (native picker:
+  // GPT-6.1 Sol, Astra, Sol, Luna) and direct OpenAI /v1/responses. Bare
+  // `gpt-6` follows the Codex CLI convention of resolving to the flagship
+  // tier (Sol). Defaults inherit the family pattern (high flagship, medium
+  // luna) pending Codex-default evidence — see the gpt6.ts descriptor
+  // comment. gpt-5.5 stays pinned: it retires from Codex sign-in 2026-10-14
+  // and codexplan already resolves to gpt-5.6-sol above.
+  'gpt-6': {
+    model: 'gpt-6-sol',
+    reasoningEffort: 'high',
+  },
+  'gpt-6-astra': {
+    model: 'gpt-6-astra',
+    reasoningEffort: 'high',
+  },
+  'gpt-6.1-sol': {
+    model: 'gpt-6.1-sol',
+    reasoningEffort: 'high',
+  },
+  'gpt-6-sol': {
+    model: 'gpt-6-sol',
+    reasoningEffort: 'high',
+  },
+  'gpt-6-luna': {
+    model: 'gpt-6-luna',
+    reasoningEffort: 'medium',
+  },
   'gpt-5.5': {
     model: 'gpt-5.5',
     reasoningEffort: 'high',
@@ -487,23 +514,33 @@ function shouldUseGithubResponsesApi(model: string): boolean {
 // gpt-5-mini, gpt-4.x, o-series, and claude-* stay on chat/completions.
 export function modelRequiresResponsesApi(model: string): boolean {
   const normalized = model.trim().toLowerCase().split('?', 1)[0] ?? ''
-  return /^gpt-5\.[4-6](?!\d)/.test(normalized) &&
+  const isGpt56Responses = /^gpt-5\.[4-6](?!\d)/.test(normalized)
+  // GPT-6 family inherits the 5.6 responses routing: same reasoning-model
+  // shape (function tools + reasoning_effort rejected on chat/completions).
+  // Verified 5.6 behavior, inherited pending live GPT-6 wire confirmation —
+  // see the gpt6.ts descriptor comment.
+  const isGpt6Responses = /^gpt-6(?:[.-]|$)/.test(normalized)
+  return (isGpt56Responses || isGpt6Responses) &&
     !GPT5_MINI_NANO_RE.test(normalized)
 }
 
 // The gpt-5 family boundary: gpt-5, gpt-5-*, gpt-5.x — without matching
 // gpt-50-style ids. Shared by supportsCodexReasoningEffort and the Codex
 // profile model gate so the family shape lives in one place.
 const GPT5_FAMILY_RE = /^gpt-5(?:[.-]|$)/
+// GPT-6 family (October 2026): same Codex-backend eligibility and effort
+// support as gpt-5. Separate pattern so the 5.x boundary stays legible.
+const GPT6_FAMILY_RE = /^gpt-6(?:[.-]|$)/
 const GPT5_MINI_NANO_RE = /(?:^|[-.])(?:mini|nano)(?:[-.]|$)/
 
 // gpt-5 family models the ChatGPT Codex backend can serve. The -mini/-nano
 // tiers are API-only (never exposed through the Codex transport), so a
 // stale gpt-5-mini pick saved under a direct-OpenAI profile must fall back
 // to the Codex profile's default rather than be sent to the backend and 400.
+// Covers gpt-6 the same way (native Codex picker serves the GPT-6 family).
 export function isCodexEligibleGpt5Model(model: string): boolean {
   const base = model.trim().toLowerCase().split('?', 1)[0] ?? ''
-  return GPT5_FAMILY_RE.test(base) && !GPT5_MINI_NANO_RE.test(base)
+  return (GPT5_FAMILY_RE.test(base) || GPT6_FAMILY_RE.test(base)) && !GPT5_MINI_NANO_RE.test(base)
 }
 
 // The responses auto-route only fires for the OpenAI first-party surface
@@ -1543,5 +1580,5 @@ export function supportsCodexReasoningEffort(model: string): boolean {
     return true
   }
 
-  return GPT5_FAMILY_RE.test(base)
+  return GPT5_FAMILY_RE.test(base) || GPT6_FAMILY_RE.test(base)
 }

src/utils/context.test.ts

@@ -458,6 +458,39 @@ test('gpt-5.6 family keeps the full window on the direct-OpenAI route', () => {
   }
 })
 
+test('gpt-6 family ships the full API spec on the Codex route (no Codex cap published)', () => {
+  process.env.CLAUDE_CODE_USE_OPENAI = '1'
+  process.env.OPENAI_BASE_URL = 'https://chatgpt.com/backend-api/codex'
+  delete process.env.CLAUDE_CODE_MAX_OUTPUT_TOKENS
+  delete process.env.OPENAI_MODEL
+
+  // Unlike the 5.6 family (pinned to the verified ~272k Codex effective
+  // input boundary), GPT-6 has no published Codex-route cap — the 272k pin
+  // is verified ONLY for 5.6 (openai/codex#33972). The gpt6.ts descriptors
+  // carry the full 1.05M spec with a dated comment + first-500 monitoring
+  // flag; if the Codex backend enforces a lower cap, pin them then.
+  for (const model of ['gpt-6-astra', 'gpt-6.1-sol', 'gpt-6-sol', 'gpt-6-luna']) {
+    expect(getContextWindowForModel(model)).toBe(1_050_000)
+    expect(getModelMaxOutputTokens(model)).toEqual({
+      default: 128_000,
+      upperLimit: 128_000,
+    })
+  }
+})
+
+test('gpt-6 family keeps the full window on the direct-OpenAI route', () => {
+  process.env.CLAUDE_CODE_USE_OPENAI = '1'
+  delete process.env.OPENAI_BASE_URL
+  delete process.env.CLAUDE_CODE_MAX_OUTPUT_TOKENS
+  delete process.env.OPENAI_MODEL
+
+  // The openai-route catalog entries preserve the true 1.05M window,
+  // same as the 5.6 family.
+  for (const model of ['gpt-6-astra', 'gpt-6.1-sol', 'gpt-6-sol', 'gpt-6-luna']) {
+    expect(getContextWindowForModel(model)).toBe(1_050_000)
+  }
+})
+
 test('gpt-5.4 family uses provider-specific context and output caps', () => {
   process.env.CLAUDE_CODE_USE_OPENAI = '1'
   delete process.env.CLAUDE_CODE_MAX_OUTPUT_TOKENS