fix(server): align Codex Fast with documented model support (#4640)

The prefix allowlist hid Fast for GPT-6 Astra and exposed it for models outside documented Codex Fast support. Use explicit model IDs and avoid claiming a universal usage multiplier.

packages/server/src/server/agent/providers/codex-app-server-agent.features.test.ts

@@ -97,6 +97,88 @@ async function createConnectedSession(
 }
 
 describe("Codex app-server provider features", () => {
+  test.each([
+    "gpt-6-astra",
+    "gpt-5.6",
+    "gpt-5.6-sol",
+    "gpt-5.6-terra",
+    "gpt-5.6-luna",
+    "gpt-5.5",
+    "gpt-5.4",
+  ])("exposes and sends Fast for %s", async (model) => {
+    const { session, appServer } = await createConnectedSession({ model });
+    try {
+      expect(session.features).toContainEqual(
+        expect.objectContaining({
+          id: "fast_mode",
+          value: false,
+        }),
+      );
+      await session.setFeature?.("fast_mode", true);
+      await session.startTurn("hello");
+      await expect(appServer.waitForTurnStart()).resolves.toMatchObject({
+        model,
+        serviceTier: "fast",
+      });
+    } finally {
+      await session.close();
+    }
+  });
+
+  test.each([
+    "gpt-5.3-codex-spark",
+    "gpt-5.3-codex",
+    "gpt-5.4-mini",
+    "gpt-5.4-nano",
+    "gpt-5.5-pro",
+    "gpt-5",
+    "gpt-4.1",
+    "o3",
+    "o4-mini",
+    "gpt-6-unknown",
+  ])("does not expose or restore Fast for %s", async (model) => {
+    const { session, appServer } = await createConnectedSession({
+      model,
+      featureValues: { fast_mode: true },
+    });
+    try {
+      expect(session.features.map((feature) => feature.id)).toEqual(["plan_mode"]);
+      await expect(session.setFeature?.("fast_mode", true)).rejects.toThrow(
+        `Codex fast mode is not available for model '${model}'`,
+      );
+      await session.startTurn("hello");
+      await expect(appServer.waitForTurnStart()).resolves.not.toMatchObject({
+        serviceTier: expect.anything(),
+      });
+    } finally {
+      await session.close();
+    }
+  });
+
+  test("restores Fast on Astra and preserves it when switching supported models", async () => {
+    const { session, appServer } = await createConnectedSession({
+      model: "gpt-6-astra",
+      featureValues: { fast_mode: true },
+    });
+    try {
+      expect(session.features).toContainEqual(
+        expect.objectContaining({
+          id: "fast_mode",
+          value: true,
+        }),
+      );
+      await session.setModel("gpt-5.6-sol");
+      await session.setModel("gpt-6-astra");
+      await session.startTurn("hello");
+      await expect(appServer.waitForTurnStart()).resolves.toMatchObject({
+        model: "gpt-6-astra",
+        serviceTier: "fast",
+      });
+    } finally {
+      await session.close();
+    }
+  });
+
   test("features returns fast and plan toggles when supported", async () => {
     const { session } = await createConnectedSession();
 
@@ -105,7 +187,7 @@ describe("Codex app-server provider features", () => {
         type: "toggle",
         id: "fast_mode",
         label: "Fast",
-        description: "Priority inference at 2x usage",
+        description: "Priority inference at increased usage",
         tooltip: "Toggle fast mode",
         icon: "zap",
         value: false,
@@ -129,7 +211,7 @@ describe("Codex app-server provider features", () => {
         type: "toggle",
         id: "fast_mode",
         label: "Fast",
-        description: "Priority inference at 2x usage",
+        description: "Priority inference at increased usage",
         tooltip: "Toggle fast mode",
         icon: "zap",
         value: true,
@@ -250,7 +332,7 @@ describe("Codex app-server provider features", () => {
         type: "toggle",
         id: "fast_mode",
         label: "Fast",
-        description: "Priority inference at 2x usage",
+        description: "Priority inference at increased usage",
         tooltip: "Toggle fast mode",
         icon: "zap",
         value: true,

packages/server/src/server/agent/providers/codex-feature-definitions.ts

@@ -1,12 +1,22 @@
 import type { AgentFeature, AgentFeatureToggle } from "../agent-sdk-types.js";
 
-const CODEX_FAST_MODE_SUPPORTED_MODEL_PREFIXES = ["gpt-5", "gpt-4.1", "o3", "o4-mini"] as const;
+// Codex Fast is distinct from API Priority processing. Keep model support aligned with
+// https://developers.openai.com/codex/speed and https://developers.openai.com/codex/models.
+const CODEX_FAST_MODE_SUPPORTED_MODELS = new Set([
+  "gpt-6-astra",
+  "gpt-5.6",
+  "gpt-5.6-sol",
+  "gpt-5.6-terra",
+  "gpt-5.6-luna",
+  "gpt-5.5",
+  "gpt-5.4",
+]);
 
 export const CODEX_FAST_MODE_FEATURE: Omit<AgentFeatureToggle, "value"> = {
   type: "toggle",
   id: "fast_mode",
   label: "Fast",
-  description: "Priority inference at 2x usage",
+  description: "Priority inference at increased usage",
   tooltip: "Toggle fast mode",
   icon: "zap",
 };
@@ -30,9 +40,7 @@ export function codexModelSupportsFastMode(modelId: string | null | undefined):
   if (!normalizedModelId) {
     return false;
   }
-  return CODEX_FAST_MODE_SUPPORTED_MODEL_PREFIXES.some(
-    (prefix) => normalizedModelId === prefix || normalizedModelId.startsWith(prefix),
-  );
+  return CODEX_FAST_MODE_SUPPORTED_MODELS.has(normalizedModelId);
 }
 
 export function buildCodexFeatures(input: {