fix(auth): drop the models retired for ChatGPT sign-in

packages/first-class-providers/src/providers.ts

@@ -55,10 +55,9 @@ export const OPENAI_API_MAX_COMPLETION_TOKENS_MODELS: readonly string[] = [
  * Tier A order: dual-path OpenAI, OAuth xAI, Go/Zen, Z.AI, big three, Custom.
  *
  * Canonical static registry for provider identity: base URLs, model lists,
- * and defaults live here. The Codex/xAI OAuth fallbacks stay separate — they
- * back live catalog calls, not static identity — but must agree with this
- * registry; src/provider/identity-divergence.test.ts pins that. Do not add
- * another edge list: reference this registry or the fallbacks.
+ * and defaults live here. The Codex/xAI OAuth model lists stay separate, since
+ * those sign-ins serve their own model sets and defaults. Do not add another
+ * edge list: reference this registry or those lists.
  */
 export const FIRST_CLASS_PROVIDERS: readonly FirstClassProviderDef[] = [
   {

src/auth/codex/constants.test.ts

@@ -0,0 +1,27 @@
+import { describe, expect, test } from "bun:test";
+
+import { CODEX_DEFAULT_MODELS } from "./constants.js";
+
+describe("CODEX_DEFAULT_MODELS", () => {
+  // Recorded from GET /codex/models for a ChatGPT Plus account (client_version
+  // 0.165.0, 2026-11-01). Nothing else pins the order or the default: the other
+  // tests read the constant.
+  test("follows the order the Codex backend listed the models in", () => {
+    expect([...CODEX_DEFAULT_MODELS]).toEqual([
+      "gpt-6.1-sol",
+      "gpt-6-astra",
+      "gpt-6.1-terra",
+      "gpt-6-sol",
+      "gpt-6-luna",
+      "gpt-5.6-sol",
+      "gpt-5.6-terra",
+      "gpt-5.6-luna",
+    ]);
+  });
+
+  test("leaves out the models OpenAI retired for ChatGPT sign-in", () => {
+    for (const retired of ["gpt-5.4", "gpt-5.4-mini", "gpt-5.5"]) {
+      expect(CODEX_DEFAULT_MODELS as readonly string[]).not.toContain(retired);
+    }
+  });
+});

src/auth/codex/constants.ts

@@ -21,19 +21,20 @@ export const CODEX_CLIENT_VERSION = "0.50.0";
 // the public Codex CLI originator.
 export const CODEX_ORIGINATOR = "codex_cli_rs";
 
-// Fallback model list, used only when the live catalog (GET /codex/models) is
-// unavailable — rate-limited it returns an empty list. The backend rotates its
-// serving set, so the live fetch is authoritative and these are just a
-// current-generation default so the picker is never empty. The first entry
-// doubles as the ChatGPT-OAuth default: it must stay the model shared with the
-// OpenAI API-key path's default in FIRST_CLASS_PROVIDERS (gpt-5.4) so both
-// auth paths serving OpenAI agree. identity-divergence.test.ts pins this.
+// The models a ChatGPT sign-in offers, in the order GET /codex/models listed
+// them for a ChatGPT Plus account (client_version 0.165.0, 2026-11-01); the
+// first entry is the ChatGPT default. Availability varies by plan and rollout,
+// so other accounts may be served a different set. gpt-5.4, gpt-5.4-mini and
+// gpt-5.5 are left out because OpenAI retired them from Codex with ChatGPT
+// sign-in. API-key sessions are not affected, so the API-key path keeps its
+// own list and default.
 export const CODEX_DEFAULT_MODELS = [
-  "gpt-5.4",
-  "gpt-5.5",
+  "gpt-6.1-sol",
   "gpt-6-astra",
+  "gpt-6.1-terra",
+  "gpt-6-sol",
+  "gpt-6-luna",
   "gpt-5.6-sol",
   "gpt-5.6-terra",
   "gpt-5.6-luna",
-  "gpt-5.4-mini",
 ] as const;

src/config/codex-providers.test.ts

@@ -22,8 +22,8 @@ describe("codex provider naming", () => {
   });
 });
 
-// The default-model agreement is pinned without literals in
-// provider/identity-divergence.test.ts — no second literal re-pin here.
+// The projected models and default are pinned against CODEX_DEFAULT_MODELS in
+// provider/identity-divergence.test.ts; no literal re-pin here.
 describe("codexProvidersAsSettings", () => {
   test("projects profiles into provider settings seeded with the access token", () => {
     const profiles: CodexProfile[] = [

src/provider/identity-divergence.test.ts

@@ -56,20 +56,6 @@ describe("provider identity divergence", () => {
     resetGoModelDiscoveryForTests();
   });
 
-  test("OpenAI API-key default and ChatGPT-OAuth fallback default agree", () => {
-    const apiPath = firstClassProviderById("openai")?.paths?.find(
-      (p) => p.id === "api",
-    );
-    if (apiPath?.defaultModel === undefined)
-      throw new Error("Expected the OpenAI API path to declare a defaultModel");
-    // Both auth paths serve OpenAI models, so the shared model must lead both
-    // lists — the fallback default is models[0] by construction.
-    const codexFallbacks: readonly string[] = CODEX_DEFAULT_MODELS;
-    const codexDefault: string | undefined = codexFallbacks[0];
-    expect(codexDefault).toBe(apiPath.defaultModel);
-    expect(codexFallbacks).toContain(apiPath.defaultModel);
-  });
-
   test("OpenAI API-key entry is self-consistent", () => {
     const apiPath = firstClassProviderById("openai")?.paths?.find(
       (p) => p.id === "api",

src/provider/reasoning-effort.test.ts

@@ -1,4 +1,5 @@
 import { afterEach, describe, test, expect } from "bun:test";
+import { CODEX_DEFAULT_MODELS } from "../auth/codex/constants.js";
 import {
   isReasoningEffort,
   normalizeProviderEfforts,
@@ -54,32 +55,53 @@ describe("supportedEfforts", () => {
     ]);
   });
 
-  test("gpt-6-astra takes low through max on both API and Codex paths", () => {
+  test("gpt-6-astra takes low through max on the API path", () => {
     expect(supportedEfforts("gpt-6-astra")).toEqual([
       "low",
       "medium",
       "high",
       "xhigh",
       "max",
     ]);
-    expect(supportedEfforts("gpt-6-astra", undefined, true)).toEqual([
-      "low",
-      "medium",
-      "high",
-      "xhigh",
-      "max",
-    ]);
   });
 
-  test("gpt-5.6 family additionally takes max and ultra on the codex backend", () => {
-    expect(supportedEfforts("gpt-5.6-sol", undefined, true)).toEqual([
-      "low",
-      "medium",
-      "high",
-      "xhigh",
-      "max",
-      "ultra",
-    ]);
+  test("ChatGPT models the backend serves up to ultra offer max and ultra", () => {
+    for (const model of [
+      "gpt-6.1-sol",
+      "gpt-6-astra",
+      "gpt-6.1-terra",
+      "gpt-6-sol",
+      "gpt-5.6-sol",
+      "gpt-5.6-terra",
+    ]) {
+      expect(supportedEfforts(model, undefined, true)).toEqual([
+        "low",
+        "medium",
+        "high",
+        "xhigh",
+        "max",
+        "ultra",
+      ]);
+    }
+  });
+
+  test("ChatGPT Luna models stop at max", () => {
+    for (const model of ["gpt-6-luna", "gpt-5.6-luna"]) {
+      const efforts = supportedEfforts(model, undefined, true);
+      expect(efforts).toEqual(["low", "medium", "high", "xhigh", "max"]);
+      expect(efforts).not.toContain("ultra");
+    }
+  });
+
+  test("every built-in ChatGPT model has an assigned effort ladder", () => {
+    // A model added to the fallback list without a ladder would silently
+    // stop at xhigh. A model the backend really serves only up to xhigh
+    // belongs here, so adding one is a decision rather than an accident.
+    const servedOnlyUpToXhigh: readonly string[] = [];
+    for (const model of CODEX_DEFAULT_MODELS) {
+      if (servedOnlyUpToXhigh.includes(model)) continue;
+      expect(supportedEfforts(model, undefined, true)).toContain("max");
+    }
   });
 
   test("the model name alone does not imply codex levels", () => {

src/provider/reasoning-effort.ts

@@ -56,19 +56,30 @@ const FULL_EFFORT_MODELS: readonly string[] = [
   "gpt-5.1-codex-max",
 ];
 
-// Codex backend models take low/medium/high/xhigh — no `minimal`, no `none`.
+// ChatGPT (Codex backend) models not listed below: low/medium/high/xhigh,
+// no `minimal`, no `none`.
 const CODEX_EFFORTS: readonly ReasoningEffort[] = [
   "low",
   "medium",
   "high",
   "xhigh",
 ];
 
-// gpt-5.6 family adds `max` and `ultra`; older Codex models do not support
-// them.
-const MAX_EFFORT_CODEX_MODELS: readonly string[] = [
+// Per-model levels the Codex backend lists (`supported_reasoning_levels` in
+// GET /codex/models, client_version 0.165.0, 2026-10-28). Exact ids rather
+// than a name rule: an id not seen yet gets the conservative fallback above
+// instead of a guessed tier, because the adapter does not clamp and a level
+// the backend does not serve breaks the turn.
+const ULTRA_EFFORT_CODEX_MODELS: readonly string[] = [
+  "gpt-6.1-sol",
+  "gpt-6-astra",
+  "gpt-6.1-terra",
+  "gpt-6-sol",
   "gpt-5.6-sol",
   "gpt-5.6-terra",
+];
+const MAX_EFFORT_CODEX_MODELS: readonly string[] = [
+  "gpt-6-luna",
   "gpt-5.6-luna",
 ];
 
@@ -96,8 +107,8 @@ function isMuseSparkModel(model: string): boolean {
 const GROK_46_AND_47_EFFORTS: readonly ReasoningEffort[] = CODEX_EFFORTS;
 const GROK_46_AND_47_MODELS: readonly string[] = ["grok-4.6", "grok-4.7"];
 
-// GPT-6 Astra accepts low–max on both surfaces; not ultra (gpt-5.6
-// Codex-only), minimal, or none.
+// GPT-6 Astra on the API path accepts low–max; not ultra, minimal, or none.
+// On ChatGPT sign-in it follows the served-levels lists above.
 const GPT6_ASTRA_EFFORTS: readonly ReasoningEffort[] = [
   "low",
   "medium",
@@ -152,14 +163,17 @@ export function supportedEfforts(
   if (providerEfforts.length > 0) {
     return [...providerEfforts];
   }
-  if (model === "gpt-6-astra") {
-    return [...GPT6_ASTRA_EFFORTS];
-  }
   if (isCodex) {
+    if (ULTRA_EFFORT_CODEX_MODELS.includes(model)) {
+      return [...CODEX_EFFORTS, "max", "ultra"];
+    }
     return MAX_EFFORT_CODEX_MODELS.includes(model)
-      ? [...CODEX_EFFORTS, "max", "ultra"]
+      ? [...CODEX_EFFORTS, "max"]
       : [...CODEX_EFFORTS];
   }
+  if (model === "gpt-6-astra") {
+    return [...GPT6_ASTRA_EFFORTS];
+  }
   if (FULL_EFFORT_MODELS.includes(model)) {
     return ["none", ...DEFAULT_EFFORTS, "xhigh"];
   }