Model: add strict gpt-5.3-codex fallback for OpenAI Codex (fixes #9989) (#9995)

* Model: allow forward-compatible OpenAI Codex GPT-5 IDs

* Model: scope Codex fallback to gpt-5.3-codex

* fix: reorder codex fallback before providerCfg, add ordering test, changelog (#9989) (thanks @w1kke)

---------

Co-authored-by: Robin <4robinlehmann@gmail.com>

src/agents/pi-embedded-runner/model.test.ts

@@ -1,4 +1,4 @@
-import { describe, expect, it, vi } from "vitest";
+import { beforeEach, describe, expect, it, vi } from "vitest";
 
 vi.mock("../pi-model-discovery.js", () => ({
   discoverAuthStorage: vi.fn(() => ({ mocked: true })),
@@ -6,6 +6,7 @@ vi.mock("../pi-model-discovery.js", () => ({
 }));
 
 import type { OpenClawConfig } from "../../config/config.js";
+import { discoverModels } from "../pi-model-discovery.js";
 import { buildInlineProviderModels, resolveModel } from "./model.js";
 
 const makeModel = (id: string) => ({
@@ -18,6 +19,12 @@ const makeModel = (id: string) => ({
   maxTokens: 1,
 });
 
+beforeEach(() => {
+  vi.mocked(discoverModels).mockReturnValue({
+    find: vi.fn(() => null),
+  } as unknown as ReturnType<typeof discoverModels>);
+});
+
 describe("buildInlineProviderModels", () => {
   it("attaches provider ids to inline models", () => {
     const providers = {
@@ -127,4 +134,74 @@ describe("resolveModel", () => {
     expect(result.model?.provider).toBe("custom");
     expect(result.model?.id).toBe("missing-model");
   });
+
+  it("builds an openai-codex fallback for gpt-5.3-codex", () => {
+    const templateModel = {
+      id: "gpt-5.2-codex",
+      name: "GPT-5.2 Codex",
+      provider: "openai-codex",
+      api: "openai-codex-responses",
+      baseUrl: "https://chatgpt.com/backend-api",
+      reasoning: true,
+      input: ["text", "image"] as const,
+      cost: { input: 1.75, output: 14, cacheRead: 0.175, cacheWrite: 0 },
+      contextWindow: 272000,
+      maxTokens: 128000,
+    };
+
+    vi.mocked(discoverModels).mockReturnValue({
+      find: vi.fn((provider: string, modelId: string) => {
+        if (provider === "openai-codex" && modelId === "gpt-5.2-codex") {
+          return templateModel;
+        }
+        return null;
+      }),
+    } as unknown as ReturnType<typeof discoverModels>);
+
+    const result = resolveModel("openai-codex", "gpt-5.3-codex", "/tmp/agent");
+
+    expect(result.error).toBeUndefined();
+    expect(result.model).toMatchObject({
+      provider: "openai-codex",
+      id: "gpt-5.3-codex",
+      api: "openai-codex-responses",
+      baseUrl: "https://chatgpt.com/backend-api",
+      reasoning: true,
+      contextWindow: 272000,
+      maxTokens: 128000,
+    });
+  });
+
+  it("keeps unknown-model errors for non-gpt-5 openai-codex ids", () => {
+    const result = resolveModel("openai-codex", "gpt-4.1-mini", "/tmp/agent");
+    expect(result.model).toBeUndefined();
+    expect(result.error).toBe("Unknown model: openai-codex/gpt-4.1-mini");
+  });
+
+  it("uses codex fallback even when openai-codex provider is configured", () => {
+    // This test verifies the ordering: codex fallback must fire BEFORE the generic providerCfg fallback.
+    // If ordering is wrong, the generic fallback would use api: "openai-responses" (the default)
+    // instead of "openai-codex-responses".
+    const cfg: OpenClawConfig = {
+      models: {
+        providers: {
+          "openai-codex": {
+            baseUrl: "https://custom.example.com",
+            // No models array, or models without gpt-5.3-codex
+          },
+        },
+      },
+    } as OpenClawConfig;
+
+    vi.mocked(discoverModels).mockReturnValue({
+      find: vi.fn(() => null),
+    } as unknown as ReturnType<typeof discoverModels>);
+
+    const result = resolveModel("openai-codex", "gpt-5.3-codex", "/tmp/agent", cfg);
+
+    expect(result.error).toBeUndefined();
+    expect(result.model?.api).toBe("openai-codex-responses");
+    expect(result.model?.id).toBe("gpt-5.3-codex");
+    expect(result.model?.provider).toBe("openai-codex");
+  });
 });