fix(codex): expose native model ids in catalog (#2012)

Integrated into release/v3.8.0

open-sse/services/model.ts

@@ -74,6 +74,7 @@ for (const [aliasOrId, models] of Object.entries(PROVIDER_MODELS)) {
 }
 const KNOWN_MODEL_IDS = new Set(MODEL_TO_PROVIDERS.keys());
 const CODEX_PREFERRED_UNPREFIXED_MODELS = new Set(["gpt-5.5"]);
+export const CODEX_NATIVE_UNPREFIXED_MODELS = new Set(["codex-auto-review"]);
 
 /**
  * Resolve provider alias to provider ID
@@ -279,6 +280,14 @@ function resolveModelByProviderInference(modelId, extendedContext) {
 
   const nonOpenAIProviders = providers.filter((p) => p !== "openai");
 
+  if (CODEX_NATIVE_UNPREFIXED_MODELS.has(modelId)) {
+    return {
+      provider: "codex",
+      model: modelId,
+      extendedContext,
+    };
+  }
+
   if (providers.includes("codex") && CODEX_PREFERRED_UNPREFIXED_MODELS.has(modelId)) {
     return {
       provider: "codex",

src/app/api/v1/models/catalog.ts

@@ -16,6 +16,7 @@ import { getAllModerationModels } from "@omniroute/open-sse/config/moderationReg
 import { getAllVideoModels } from "@omniroute/open-sse/config/videoRegistry.ts";
 import { getAllMusicModels } from "@omniroute/open-sse/config/musicRegistry.ts";
 import { REGISTRY } from "@omniroute/open-sse/config/providerRegistry.ts";
+import { CODEX_NATIVE_UNPREFIXED_MODELS } from "@omniroute/open-sse/services/model.ts";
 import { getAllSyncedAvailableModels } from "@/lib/db/models";
 import { getCompatibleFallbackModels } from "@/lib/providers/managedAvailableModels";
 import { hasEligibleConnectionForModel } from "@/domain/connectionModelRules";
@@ -403,6 +404,33 @@ export async function getUnifiedModelsResponse(
       }
     }
 
+    for (const modelId of CODEX_NATIVE_UNPREFIXED_MODELS) {
+      if (!providerSupportsModel("codex", modelId)) continue;
+      if (getModelIsHidden("codex", modelId)) continue;
+
+      const alias = providerIdToAlias.codex || "cx";
+      const aliasId = `${alias}/${modelId}`;
+      const providerIdModel = `codex/${modelId}`;
+      const entries = [
+        { id: aliasId, parent: null },
+        { id: providerIdModel, parent: aliasId },
+        { id: modelId, parent: providerIdModel },
+      ];
+
+      for (const entry of entries) {
+        if (models.some((existingModel) => existingModel.id === entry.id)) continue;
+        models.push({
+          id: entry.id,
+          object: "model",
+          created: timestamp,
+          owned_by: "codex",
+          permission: [],
+          root: modelId,
+          parent: entry.parent,
+        });
+      }
+    }
+
     try {
       const syncedModelsByProvider = await getAllSyncedAvailableModels();
       for (const [providerId, syncedModels] of Object.entries(syncedModelsByProvider)) {

src/sse/handlers/chat.ts

@@ -485,7 +485,12 @@ async function handleSingleModelChat(
   isCombo: boolean = false
 ) {
   // 1. Resolve model → provider/model
-  const resolved = await resolveModelOrError(modelStr, body, clientRawRequest?.endpoint);
+  const resolved = await resolveModelOrError(
+    modelStr,
+    body,
+    clientRawRequest?.endpoint,
+    clientRawRequest?.headers
+  );
   if (resolved.error) return resolved.error;
 
   const { provider, model, sourceFormat, targetFormat, extendedContext } = resolved;

src/sse/handlers/chatHelpers.ts

@@ -43,8 +43,59 @@ const PREFERRED_BY_FAMILY: Record<string, string> = {
   mimo: "moonshot",
 };
 
-export async function resolveModelOrError(modelStr: string, body: any, endpointPath: string = "") {
+const CODEX_NATIVE_RESPONSES_MODELS = new Set(["gpt-5.5"]);
+
+function getHeaderValue(headers: Record<string, unknown> | null | undefined, name: string) {
+  if (!headers || typeof headers !== "object") return "";
+  const lowerName = name.toLowerCase();
+  for (const [key, value] of Object.entries(headers)) {
+    if (key.toLowerCase() !== lowerName) continue;
+    return Array.isArray(value) ? value.join(",") : String(value ?? "");
+  }
+  return "";
+}
+
+function isCodexNativeResponsesRequest(
+  body: any,
+  endpointPath: string,
+  headers: Record<string, unknown> | null | undefined
+) {
+  const normalizedEndpoint = String(endpointPath || "").replace(/\/+$/, "");
+  if (!/(^|\/)responses(?=\/|$)/i.test(normalizedEndpoint)) return false;
+  if (/\/responses\/compact$/i.test(normalizedEndpoint)) return true;
+
+  const userAgent = getHeaderValue(headers, "user-agent").toLowerCase();
+  if (userAgent.includes("codex")) return true;
+  if (getHeaderValue(headers, "x-codex-session-id")) return true;
+  if (getHeaderValue(headers, "x-codex-window-id")) return true;
+  if (getHeaderValue(headers, "x-codex-turn-metadata")) return true;
+
+  const metadataSource =
+    body && typeof body === "object" && body.metadata && typeof body.metadata === "object"
+      ? String(body.metadata.source || "")
+      : "";
+  return metadataSource.toLowerCase().includes("codex");
+}
+
+export async function resolveModelOrError(
+  modelStr: string,
+  body: any,
+  endpointPath: string = "",
+  requestHeaders: Record<string, unknown> | null | undefined = null
+) {
   const modelInfo = await getModelInfo(modelStr);
+  const sourceFormat = detectFormatFromEndpoint(body, endpointPath);
+
+  if (
+    modelInfo.provider === "openai" &&
+    typeof modelInfo.model === "string" &&
+    CODEX_NATIVE_RESPONSES_MODELS.has(modelInfo.model) &&
+    sourceFormat === "openai-responses" &&
+    isCodexNativeResponsesRequest(body, endpointPath, requestHeaders)
+  ) {
+    log.info("ROUTING", `${modelStr} → codex/${modelInfo.model} (Codex native responses)`);
+    modelInfo.provider = "codex";
+  }
 
   // Forced-rewrite: codex provider doesn't serve DeepSeek/Qwen/Kimi/etc. Reroute
   // these to their canonical native provider so the request lands on the right
@@ -113,7 +164,6 @@ export async function resolveModelOrError(modelStr: string, body: any, endpointP
   }
 
   const { provider, model, extendedContext } = modelInfo;
-  const sourceFormat = detectFormatFromEndpoint(body, endpointPath);
   const providerAlias = PROVIDER_ID_TO_ALIAS[provider] || provider;
   let targetFormat = getModelTargetFormat(providerAlias, model) || getTargetFormat(provider);
   if ((modelInfo as any).apiFormat === "responses") {

tests/unit/chat-helpers.test.ts

@@ -89,6 +89,30 @@ test("resolveModelOrError rejects malformed model strings", async () => {
   assert.match(json.error.message, /Invalid model format/i);
 });
 
+test("resolveModelOrError routes Codex native compact gpt-5.5 requests to Codex", async () => {
+  const result = await resolveModelOrError(
+    "gpt-5.5",
+    { model: "gpt-5.5", input: "compact this session", reasoning: { effort: "xhigh" } },
+    "/v1/responses/compact",
+    { "user-agent": "codex-cli/0.128.0" }
+  );
+
+  assert.equal(result.provider, "codex");
+  assert.equal(result.model, "gpt-5.5");
+});
+
+test("resolveModelOrError keeps non-Codex gpt-5.5 Responses requests on OpenAI", async () => {
+  const result = await resolveModelOrError(
+    "gpt-5.5",
+    { model: "gpt-5.5", input: "hello" },
+    "/v1/responses",
+    { "user-agent": "OpenAI/Node" }
+  );
+
+  assert.equal(result.provider, "openai");
+  assert.equal(result.model, "gpt-5.5");
+});
+
 test("checkPipelineGates blocks providers with an open circuit breaker", async () => {
   const breaker = getCircuitBreaker("openai");
   breaker.state = STATE.OPEN;

tests/unit/models-catalog-route.test.ts

@@ -288,6 +288,36 @@ test("v1 models catalog exposes refreshed GitHub Copilot aliases and drops retir
   );
 });
 
+test("v1 models catalog exposes bare Codex-preferred IDs for native Codex clients", async () => {
+  await seedConnection("codex", {
+    authType: "oauth",
+    name: "codex-native",
+    apiKey: null,
+    accessToken: "codex-access",
+  });
+
+  const response = await v1ModelsCatalog.getUnifiedModelsResponse(
+    new Request("http://localhost/api/v1/models")
+  );
+  const body = (await response.json()) as any;
+  const getModel = (id: string) => body.data.find((item) => item.id === id);
+
+  assert.equal(response.status, 200);
+  const modelId = "codex-auto-review";
+  const bareModel = getModel(modelId);
+  const providerModel = getModel(`codex/${modelId}`);
+  const aliasModel = getModel(`cx/${modelId}`);
+  const openAiModel = getModel(`openai/${modelId}`);
+
+  assert.ok(bareModel, `expected bare ${modelId} model`);
+  assert.ok(providerModel, `expected codex/${modelId} model`);
+  assert.ok(aliasModel, `expected cx/${modelId} model`);
+  assert.equal(openAiModel, undefined);
+  assert.equal(bareModel.owned_by, "codex");
+  assert.equal(bareModel.parent, providerModel.id);
+  assert.equal(providerModel.parent, aliasModel.id);
+});
+
 test("v1 models catalog exposes Antigravity client-visible preview aliases instead of upstream internal IDs", async () => {
   await seedConnection("antigravity", {
     authType: "oauth",

tests/unit/plan3-p0.test.ts

@@ -28,9 +28,9 @@ test("getModelInfoCore keeps openai fallback for gpt-4o", async () => {
   assert.equal(info.model, "gpt-4o");
 });
 
-test("getModelInfoCore routes removed codex-auto-review through the default fallback", async () => {
+test("getModelInfoCore routes native codex-auto-review to codex", async () => {
   const info = await getModelInfoCore("codex-auto-review", {});
-  assert.equal(info.provider, "openai");
+  assert.equal(info.provider, "codex");
   assert.equal(info.model, "codex-auto-review");
 });