feat(providers): refresh Z.ai Web models and browser transport (#12524)

Validado em lote numa worktree combinada com #12524, #12538, #12277 e #12367 sobre o tip de release/v3.8.51: os quatro boardaram sem conflito (áreas disjuntas — zai-web, nvidia, clova, cursor/devin/fable). typecheck:core limpo, check:provider-consistency OK (272 entradas REGISTRY, 355 providers canônicos), check:known-symbols OK, e 305/305 nos testes tocados pelos quatro PRs. Os IDs de modelo adicionados foram conferidos individualmente. Obrigado, @backryun.
2026-09-08 07:52:12 +03:00 · 2026-09-03 19:49:42 +09:00
parent 2a6d45abea
commit 032adb0809
10 changed files with 237 additions and 186 deletions
--- a/open-sse/config/providers/registry/zai-web/index.ts
+++ b/open-sse/config/providers/registry/zai-web/index.ts
@@ -14,30 +14,27 @@ export const zai_webProvider: RegistryEntry = {
  // Z.ai's visible "Tools" switch enables its internal VLM/MCP tools. It does
  // not accept caller-supplied OpenAI `tools`, which remains disabled here.
  models: [
+    {
+      id: "glm-5.3-flash",
+      name: "GLM-5.3-Flash",
+      toolCalling: false,
+      supportsReasoning: true,
+      supportedThinkingEfforts: ["low", "high", "max"],
+      supportsVision: true,
+    },
+    {
+      id: "glm-5.3",
+      name: "GLM-5.3",
+      toolCalling: false,
+      supportsReasoning: true,
+      supportedThinkingEfforts: ["low", "high", "max"],
+    },
    {
      id: "glm-5.2",
      name: "GLM-5.2",
      toolCalling: false,
      supportsReasoning: true,
-    },
-    {
-      id: "GLM-5.1",
-      name: "GLM-5.1",
-      toolCalling: false,
-      supportsReasoning: true,
-    },
-    {
-      id: "GLM-5-Turbo",
-      name: "GLM-5-Turbo",
-      toolCalling: false,
-      supportsReasoning: true,
-    },
-    {
-      id: "GLM-5v-Turbo",
-      name: "GLM-5V-Turbo",
-      toolCalling: false,
-      supportsReasoning: true,
-      supportsVision: true,
+      supportedThinkingEfforts: ["high", "max"],
    },
  ],
 };
--- a/open-sse/executors/zai-web.ts
+++ b/open-sse/executors/zai-web.ts
@@ -5,8 +5,9 @@
 * browser-issued CAPTCHA proof for chat completions. The browser transport is
 * the default; callers with a short-lived proof can use the direct HTTP path.
 *
- * Completions go to /api/v2/chat/completions; the older unversioned
- * /api/chat/completions path is stale and 404s model-independently (#8014).
+ * Completions go to /api/v2/chat/completions. Z.ai's CAPTCHA rejects true
+ * headless Chromium with F001, so the browser transport uses off-screen headed
+ * Chromium while retaining the shared browser pool.
 */
 import { createHash, randomUUID } from "node:crypto";
 import { BaseExecutor, type ExecuteInput } from "./base.ts";
@@ -67,6 +68,7 @@ export {
  parseZaiFrontendVersion,
  resolveZaiThinkingConfig,
  resolveZaiVlmConfig,
+  zaiUpstreamModelId,
 } from "./zai-web/protocol.ts";
 export type {
  ZaiModelCapabilities,
@@ -174,6 +176,7 @@ function buildZaiBrowserChatOptions(input: {
    userAgent: ZAI_USER_AGENT,
    locale: "en-US",
    timezone: "Asia/Seoul",
+    headless: false,
    inputSelector: "#chat-input",
    submitButtonSelector: '[aria-label="Send Message"] button:not([disabled])',
    submitButtonMode: "dom",
@@ -243,7 +246,7 @@ function resolveZaiRequest(
  const modelId = (bodyObj.model as string) || model || ZAI_DEFAULT_MODEL;
  if (imageUrls.length > 0 && !getZaiModelCapabilities(modelId).vision) {
    return fail(
-      `Z.ai model ${unprefixedModelId(modelId)} does not accept image input; use GLM-5V-Turbo.`
+      `Z.ai model ${unprefixedModelId(modelId)} does not accept image input; use GLM-5.3-Flash.`
    );
  }

--- a/open-sse/executors/zai-web/browserAutomation.ts
+++ b/open-sse/executors/zai-web/browserAutomation.ts
@@ -23,14 +23,17 @@ async function runStage(name: string, action: () => Promise<void>): Promise<void
 async function selectZaiBrowserModel(page: Page, modelName: string): Promise<void> {
  const selector = page.locator('[aria-label="Select a model"]').first();
  await selector.waitFor({ state: "visible", timeout: 10_000 });
-  if ((await selector.innerText()).includes(modelName)) return;
+  if ((await selector.getByText(modelName, { exact: true }).count()) > 0) return;

  // The landing-page hero animation can remain above the already-visible
  // selector and make coordinate-based clicks time out.
  await selector.evaluate((element) => (element as HTMLElement).click());
  const menu = page.locator('[role="menu"]').filter({ hasText: modelName }).first();
  await menu.waitFor({ state: "visible", timeout: 5_000 });
-  const modelButton = menu.locator("button").filter({ hasText: modelName }).first();
+  const modelButton = menu
+    .getByText(modelName, { exact: true })
+    .first()
+    .locator("xpath=ancestor::button[1]");
  await modelButton.evaluate((element) => (element as HTMLElement).click());
  await page
    .locator('[aria-label="Select a model"]')
@@ -73,13 +76,13 @@ async function setZaiBrowserWebSearch(page: Page, enabled: boolean): Promise<voi
  }
 }

-/** Pick the High/Max effort button inside an already-open Deep Think menu. */
+/** Pick the Low/High/Max effort button inside an already-open Deep Think menu. */
 async function selectZaiBrowserEffortLevel(
  menu: ReturnType<Page["locator"]>,
  effort: ZaiThinkingConfig["effort"]
 ): Promise<void> {
  const effortButton = menu.locator("button").filter({
-    hasText: effort === "high" ? "High" : "Max",
+    hasText: effort === "low" ? "Low" : effort === "high" ? "High" : "Max",
  });
  if ((await effortButton.getAttribute("data-selected")) === "true") return;
  await runStage(`select ${effort}`, () =>
--- a/open-sse/executors/zai-web/protocol.ts
+++ b/open-sse/executors/zai-web/protocol.ts
@@ -7,8 +7,8 @@ import { normalizeCookie, sanitizeErrorMessage } from "../../utils/error.ts";
 export const ZAI_BASE_URL = "https://chat.z.ai";
 export const ZAI_NEW_CHAT_URL = `${ZAI_BASE_URL}/api/v1/chats/new`;
 export const ZAI_CHAT_URL = `${ZAI_BASE_URL}/api/v2/chat/completions`;
-export const ZAI_DEFAULT_MODEL = "GLM-5.1";
-export const ZAI_DEFAULT_FE_VERSION = "prod-fe-1.1.79";
+export const ZAI_DEFAULT_MODEL = "glm-5.3";
+export const ZAI_DEFAULT_FE_VERSION = "prod-fe-1.1.92";
 export const ZAI_USER_AGENT =
  "Mozilla/5.0 (Macintosh; Intel Mac OS X 10_15_7) AppleWebKit/537.36 (KHTML, like Gecko) Chrome/150.0.0.0 Safari/537.36";
 export const ZAI_FE_VERSION_CACHE_TTL_MS = 15 * 60 * 1000;
@@ -21,7 +21,7 @@ export interface NewChatRequest {
  userMessageId: string;
 }

-export type ZaiReasoningEffort = "high" | "max";
+export type ZaiReasoningEffort = "low" | "high" | "max";

 export interface ZaiThinkingConfig {
  enabled: boolean;
@@ -61,12 +61,23 @@ const NO_ZAI_MODEL_CAPABILITIES: ZaiModelCapabilities = Object.freeze({
 });

 /**
- * Verified against chat.z.ai/api/models (prod-fe-1.1.79).
+ * Verified against chat.z.ai/api/models (prod-fe-1.1.92).
 * `returnFc` is the site's internal function-call result capability; it is
 * distinct from accepting caller-supplied OpenAI `tools`.
 */
 const ZAI_MODEL_CAPABILITIES: Record<string, ZaiModelCapabilities> = {
-  "glm-5.2": {
+  "glm-5.3-flash": {
+    mcp: false,
+    reasoningEffort: true,
+    returnFc: true,
+    thinking: true,
+    vision: true,
+    vlmTools: false,
+    vlmWebSearch: false,
+    vlmWebsiteMode: false,
+    webSearch: true,
+  },
+  "glm-5.3": {
    mcp: true,
    reasoningEffort: true,
    returnFc: true,
@@ -77,9 +88,9 @@ const ZAI_MODEL_CAPABILITIES: Record<string, ZaiModelCapabilities> = {
    vlmWebsiteMode: false,
    webSearch: true,
  },
-  "glm-5.1": {
+  "glm-5.2": {
    mcp: true,
-    reasoningEffort: false,
+    reasoningEffort: true,
    returnFc: true,
    thinking: true,
    vision: false,
@@ -88,28 +99,6 @@ const ZAI_MODEL_CAPABILITIES: Record<string, ZaiModelCapabilities> = {
    vlmWebsiteMode: false,
    webSearch: true,
  },
-  "glm-5-turbo": {
-    mcp: true,
-    reasoningEffort: false,
-    returnFc: true,
-    thinking: true,
-    vision: false,
-    vlmTools: false,
-    vlmWebSearch: false,
-    vlmWebsiteMode: false,
-    webSearch: true,
-  },
-  "glm-5v-turbo": {
-    mcp: false,
-    reasoningEffort: false,
-    returnFc: true,
-    thinking: true,
-    vision: true,
-    vlmTools: true,
-    vlmWebSearch: true,
-    vlmWebsiteMode: true,
-    webSearch: true,
-  },
 };

 export function asRecord(value: unknown): Record<string, unknown> | null {
@@ -136,6 +125,7 @@ export function describeZaiBrowserFailure(result: {
  status: number;
  body: Buffer;
  observedPostUrls?: string[];
+  observedPostResponses?: Array<{ url: string; status: number }>;
  timing: { captureResponseMs: number; totalMs: number };
 }): string {
  const status = result.status > 0 ? String(result.status) : "no matching response";
@@ -144,10 +134,16 @@ export function describeZaiBrowserFailure(result: {
    result.observedPostUrls && result.observedPostUrls.length > 0
      ? ` Observed POST targets: ${result.observedPostUrls.join(", ")}.`
      : "";
+  const observedResponses =
+    result.observedPostResponses && result.observedPostResponses.length > 0
+      ? ` Observed POST responses: ${result.observedPostResponses
+          .map(({ url, status }) => `${url} [${status}]`)
+          .join(", ")}.`
+      : "";
  const detail =
    browserFailureDetail(result.body) ||
    (result.status === 0
-      ? `The page did not issue the expected authenticated chat completion request.${observed}`
+      ? `The page did not issue the expected authenticated chat completion request.${observed}${observedResponses}`
      : "The browser response body was empty.");
  return `Z.ai browser transport failed (${status}; ${timing}): ${detail}`;
 }
@@ -307,17 +303,27 @@ export function unprefixedModelId(modelId: string): string {
  return modelId.trim().split("/").at(-1) || modelId.trim();
 }

-export function browserModelName(modelId: string): string {
+/** Map OmniRoute's public Flash id to the opaque id used by chat.z.ai's wire API. */
+export function zaiUpstreamModelId(modelId: string): string {
  const unprefixed = unprefixedModelId(modelId);
-  if (unprefixed.toLowerCase() === "glm-5.2") return "GLM-5.2";
-  if (unprefixed.toLowerCase() === "glm-5v-turbo") return "GLM-5V-Turbo";
-  return unprefixed;
+  return unprefixed.toLowerCase() === "glm-5.3-flash" ? "x-preview-l" : unprefixed;
+}
+
+function zaiCapabilityModelId(modelId: string): string {
+  const unprefixed = unprefixedModelId(modelId).toLowerCase();
+  return unprefixed === "x-preview-l" ? "glm-5.3-flash" : unprefixed;
+}
+
+export function browserModelName(modelId: string): string {
+  const normalized = zaiCapabilityModelId(modelId);
+  if (normalized === "glm-5.3-flash") return "GLM-5.3-Flash";
+  if (normalized === "glm-5.3") return "GLM-5.3";
+  if (normalized === "glm-5.2") return "GLM-5.2";
+  return unprefixedModelId(modelId);
 }

 export function getZaiModelCapabilities(modelId: string): ZaiModelCapabilities {
-  return (
-    ZAI_MODEL_CAPABILITIES[unprefixedModelId(modelId).toLowerCase()] ?? NO_ZAI_MODEL_CAPABILITIES
-  );
+  return ZAI_MODEL_CAPABILITIES[zaiCapabilityModelId(modelId)] ?? NO_ZAI_MODEL_CAPABILITIES;
 }

 function getFeatureOption(body: Record<string, unknown>, key: string): unknown {
@@ -325,7 +331,7 @@ function getFeatureOption(body: Record<string, unknown>, key: string): unknown {
  return asRecord(body.features)?.[key];
 }

-/** Resolve each model's Deep Think control; only GLM-5.2 accepts High/Max effort. */
+/** Resolve each model's Deep Think control using its currently exposed effort vocabulary. */
 export function resolveZaiThinkingConfig(
  modelId: string,
  body: Record<string, unknown>
@@ -339,19 +345,26 @@ export function resolveZaiThinkingConfig(
      : typeof reasoning?.effort === "string"
        ? reasoning.effort.trim().toLowerCase()
        : "";
-  const disabled = body.enable_thinking === false || rawEffort === "none" || rawEffort === "off";
+  const supportsLowEffort = zaiCapabilityModelId(modelId) !== "glm-5.2";
  const effort: ZaiReasoningEffort =
-    rawEffort === "low" || rawEffort === "medium" || rawEffort === "high" ? "high" : "max";
+    rawEffort === "low" && supportsLowEffort
+      ? "low"
+      : rawEffort === "low" || rawEffort === "medium" || rawEffort === "high"
+        ? "high"
+        : "max";

  return {
    supported,
-    enabled: supported && !disabled,
+    // The current GLM-5.3/5.2 consumer models expose effort selection but no
+    // non-thinking mode. Keep Deep Think enabled even when a generic client
+    // sends an off/none compatibility value.
+    enabled: supported,
    effort,
    effortSupported: capabilities.reasoningEffort,
  };
 }

-/** Resolve GLM-5V-Turbo's visible Web Search and Tools controls. */
+/** Resolve the selected model's visible Web Search and Tools controls. */
 export function resolveZaiVlmConfig(modelId: string, body: Record<string, unknown>): ZaiVlmConfig {
  const capabilities = getZaiModelCapabilities(modelId);
  const toolsOption = getFeatureOption(body, "vlm_tools_enable");
@@ -425,7 +438,7 @@ export function buildZaiCompletionUrl(input: {
    hostname: "chat.z.ai",
    protocol: "https:",
    referrer: "",
-    title: "Z.ai - Advanced AI Chatbot & Agent powered by GLM-5.2",
+    title: "Z.ai - Advanced AI Chatbot & Agent powered by GLM-5.3",
    timezone_offset: "0",
    local_time: now.toISOString(),
    utc_time: now.toUTCString(),
@@ -448,13 +461,14 @@ export function buildZaiNewChatBody(
 ): NewChatRequest {
  const prompt = latestUserPrompt(messages);
  const userMessageId = randomUUID();
+  const upstreamModelId = zaiUpstreamModelId(modelId);
  return {
    userMessageId,
    payload: {
      chat: {
        id: "",
        title: "New Chat",
-        models: [modelId],
+        models: [upstreamModelId],
        params: {},
        history: {
          messages: {
@@ -465,7 +479,7 @@ export function buildZaiNewChatBody(
              role: "user",
              content: prompt,
              timestamp: Math.floor(Date.now() / 1000),
-              models: [modelId],
+              models: [upstreamModelId],
            },
          },
          currentId: userMessageId,
@@ -530,7 +544,7 @@ export function buildZaiRequestBody(input: {
  }
  return {
    stream: true,
-    model: input.modelId,
+    model: zaiUpstreamModelId(input.modelId),
    messages: foldMessages(input.messages),
    signature_prompt: input.prompt,
    params,
--- a/open-sse/services/browserBackedChat.ts
+++ b/open-sse/services/browserBackedChat.ts
@@ -236,6 +236,7 @@ export async function browserBackedChat(
    userAgent,
    locale,
    timezone,
+    headless,
    inputSelector,
    submitButtonSelector,
    submitButtonMode = "playwright",
@@ -257,11 +258,13 @@ export async function browserBackedChat(
    userAgent,
    locale,
    timezone,
+    headless,
  });
  const acquireContextMs = Date.now() - tAcquireStart;

  const page = await openPage(pooled);
  const observedPostUrls: string[] = [];
+  const observedPostResponses: Array<{ url: string; status: number }> = [];
  page.on("request", (request) => {
    if (request.method() !== "POST") return;
    try {
@@ -273,6 +276,19 @@ export async function browserBackedChat(
      // Ignore malformed/non-HTTP request URLs.
    }
  });
+  page.on("response", (response) => {
+    if (response.request().method() !== "POST") return;
+    try {
+      const url = new URL(response.url());
+      if (!url.hostname.endsWith(chatUrlMatchDomain)) return;
+      observedPostResponses.push({
+        url: `${url.origin}${url.pathname}`,
+        status: response.status(),
+      });
+    } catch {
+      // Ignore malformed/non-HTTP response URLs.
+    }
+  });
  try {
    const tNavStart = Date.now();
    await withAbort(
@@ -379,6 +395,7 @@ export async function browserBackedChat(
      body,
      isStealth: pooled.isStealth,
      observedPostUrls,
+      observedPostResponses,
      timing: {
        acquireContextMs,
        navigateMs,
@@ -404,6 +421,7 @@ export async function browserBackedChat(
      body,
      isStealth: pooled.isStealth,
      observedPostUrls,
+      observedPostResponses,
      timing: {
        acquireContextMs,
        navigateMs: 0,
--- a/open-sse/services/browserBackedChat/types.ts
+++ b/open-sse/services/browserBackedChat/types.ts
@@ -29,6 +29,8 @@ export interface BrowserBackedChatRequest {
  locale?: string;
  /** Browser IANA timezone. Defaults to America/New_York. */
  timezone?: string;
+  /** Launch a headed browser when the provider rejects true headless mode. */
+  headless?: boolean;
  /** Selector for the provider chat input. */
  inputSelector: string;
  /** Optional selector for the provider submit button. */
@@ -64,6 +66,8 @@ export interface BrowserBackedChatResult {
  isStealth: boolean;
  /** Sanitized POST targets observed while submitting. */
  observedPostUrls?: string[];
+  /** Sanitized POST response targets and statuses observed while submitting. */
+  observedPostResponses?: Array<{ url: string; status: number }>;
  timing: {
    acquireContextMs: number;
    navigateMs: number;
--- a/tests/unit/executor-zai-web.test.ts
+++ b/tests/unit/executor-zai-web.test.ts
@@ -30,7 +30,7 @@ function installZaiFetch(
    const value = String(url);
    if (value === ZAI_HOME_URL) {
      return new Response(
-        '<script src="https://z-cdn.chatglm.cn/z-ai/frontend/prod-fe-1.1.79/assets/index.js"></script>'
+        '<script src="https://z-cdn.chatglm.cn/z-ai/frontend/prod-fe-1.1.92/assets/index.js"></script>'
      );
    }
    if (value === ZAI_NEW_CHAT_URL) {
@@ -90,13 +90,18 @@ describe("ZaiWebExecutor", () => {
      "Z.ai browser transport failed (502; capture 30001ms, total 33412ms): " +
        "browserBackedChat failed: response.body unavailable"
    );
-    assert.match(
+    assert.equal(
      mod.describeZaiBrowserFailure({
        status: 0,
        body: Buffer.alloc(0),
+        observedPostUrls: ["https://chat.z.ai/api/v1/chats/new"],
+        observedPostResponses: [{ url: "https://chat.z.ai/api/v1/chats/new", status: 200 }],
        timing: { captureResponseMs: 30_000, totalMs: 33_000 },
      }),
-      /no matching response.*did not issue the expected authenticated chat completion request/
+      "Z.ai browser transport failed (no matching response; capture 30000ms, total 33000ms): " +
+        "The page did not issue the expected authenticated chat completion request. " +
+        "Observed POST targets: https://chat.z.ai/api/v1/chats/new. " +
+        "Observed POST responses: https://chat.z.ai/api/v1/chats/new [200]."
    );
  });

@@ -128,9 +133,9 @@ describe("ZaiWebExecutor", () => {
  it("parses the deployed frontend version from the homepage asset path", () => {
    assert.equal(
      mod.parseZaiFrontendVersion(
-        "https://z-cdn.chatglm.cn/z-ai/frontend/prod-fe-1.1.79/assets/index.js"
+        "https://z-cdn.chatglm.cn/z-ai/frontend/prod-fe-1.1.92/assets/index.js"
      ),
-      "prod-fe-1.1.79"
+      "prod-fe-1.1.92"
    );
    assert.equal(mod.parseZaiFrontendVersion("<html></html>"), null);
  });
@@ -211,86 +216,87 @@ describe("ZaiWebExecutor", () => {
    ]);
  });

-  it("enables Deep Think for every public model and limits effort to GLM-5.2", () => {
-    assert.deepEqual(mod.resolveZaiThinkingConfig("glm-5.2", {}), {
+  it("maps the three public models to their current Deep Think effort vocabularies", () => {
+    assert.deepEqual(mod.resolveZaiThinkingConfig("glm-5.3-flash", { reasoning_effort: "low" }), {
      supported: true,
      enabled: true,
-      effort: "max",
+      effort: "low",
      effortSupported: true,
    });
-    assert.deepEqual(mod.resolveZaiThinkingConfig("zw/glm-5.2", { reasoning_effort: "medium" }), {
+    assert.deepEqual(mod.resolveZaiThinkingConfig("zw/glm-5.3", { reasoning_effort: "medium" }), {
      supported: true,
      enabled: true,
      effort: "high",
      effortSupported: true,
    });
-    assert.deepEqual(mod.resolveZaiThinkingConfig("glm-5.2", { reasoning: { effort: "high" } }), {
+    assert.deepEqual(mod.resolveZaiThinkingConfig("glm-5.3", { reasoning: { effort: "high" } }), {
      supported: true,
      enabled: true,
      effort: "high",
      effortSupported: true,
    });
-    assert.deepEqual(mod.resolveZaiThinkingConfig("glm-5.2", { reasoning_effort: "off" }), {
-      supported: true,
-      enabled: false,
-      effort: "max",
-      effortSupported: true,
-    });
-    assert.deepEqual(mod.resolveZaiThinkingConfig("GLM-5.1", { reasoning_effort: "max" }), {
+    assert.deepEqual(mod.resolveZaiThinkingConfig("glm-5.3", { reasoning_effort: "off" }), {
      supported: true,
      enabled: true,
      effort: "max",
-      effortSupported: false,
+      effortSupported: true,
+    });
+    assert.deepEqual(mod.resolveZaiThinkingConfig("glm-5.3", { enable_thinking: false }), {
+      supported: true,
+      enabled: true,
+      effort: "max",
+      effortSupported: true,
+    });
+    assert.deepEqual(mod.resolveZaiThinkingConfig("glm-5.2", { reasoning_effort: "low" }), {
+      supported: true,
+      enabled: true,
+      effort: "high",
+      effortSupported: true,
    });
  });

-  it("maps GLM-5V-Turbo vision and internal VLM controls from live capabilities", () => {
-    assert.deepEqual(mod.getZaiModelCapabilities("zw/GLM-5v-Turbo"), {
+  it("maps GLM-5.3-Flash vision and web controls from live capabilities", () => {
+    assert.deepEqual(mod.getZaiModelCapabilities("zw/glm-5.3-flash"), {
      mcp: false,
-      reasoningEffort: false,
+      reasoningEffort: true,
      returnFc: true,
      thinking: true,
      vision: true,
-      vlmTools: true,
-      vlmWebSearch: true,
-      vlmWebsiteMode: true,
+      vlmTools: false,
+      vlmWebSearch: false,
+      vlmWebsiteMode: false,
      webSearch: true,
    });
-    assert.deepEqual(mod.resolveZaiVlmConfig("GLM-5v-Turbo", {}), {
-      toolsEnabled: true,
-      webSearchEnabled: true,
-      websiteModeEnabled: true,
+    assert.deepEqual(mod.getZaiModelCapabilities("x-preview-l"), {
+      mcp: false,
+      reasoningEffort: true,
+      returnFc: true,
+      thinking: true,
+      vision: true,
+      vlmTools: false,
+      vlmWebSearch: false,
+      vlmWebsiteMode: false,
+      webSearch: true,
    });
-    assert.deepEqual(
-      mod.resolveZaiVlmConfig("GLM-5v-Turbo", {
-        features: {
-          vlm_tools_enable: false,
-          vlm_web_search_enable: false,
-          vlm_website_mode: false,
-        },
-      }),
-      {
-        toolsEnabled: false,
-        webSearchEnabled: false,
-        websiteModeEnabled: true,
-      }
-    );
-    assert.deepEqual(mod.resolveZaiVlmConfig("GLM-5.1", {}), {
+    assert.deepEqual(mod.resolveZaiVlmConfig("glm-5.3-flash", { web_search: true }), {
+      toolsEnabled: false,
+      webSearchEnabled: true,
+      websiteModeEnabled: false,
+    });
+    assert.deepEqual(mod.resolveZaiVlmConfig("glm-5.3", {}), {
      toolsEnabled: false,
      webSearchEnabled: false,
      websiteModeEnabled: false,
    });
-    assert.deepEqual(mod.resolveZaiVlmConfig("GLM-5.1", { web_search: true }), {
-      toolsEnabled: false,
-      webSearchEnabled: true,
-      websiteModeEnabled: false,
-    });
+    assert.equal(mod.zaiUpstreamModelId("zai-web/glm-5.3-flash"), "x-preview-l");
+    assert.equal(mod.zaiUpstreamModelId("zai-web/glm-5.3"), "glm-5.3");
+    assert.equal(mod.getZaiModelCapabilities("GLM-5.1").thinking, false);
  });

  it("returns a credential error when no session credential is provided", async () => {
    const executor = new mod.ZaiWebExecutor();
    const result = await executor.execute({
-      model: "GLM-5.1",
+      model: "glm-5.3",
      body: { messages: [{ role: "user", content: "hi" }] },
      stream: false,
      credentials: { apiKey: "" },
@@ -313,7 +319,6 @@ describe("ZaiWebExecutor", () => {
    try {
      const executor = new mod.ZaiWebExecutor();
      const result = await executor.execute({
-        model: "glm-5.2",
        body: { messages: [{ role: "user", content: "hi" }] },
        stream: false,
        credentials: { apiKey: TEST_TOKEN },
@@ -322,8 +327,10 @@ describe("ZaiWebExecutor", () => {

      const completion = await result.response.json();
      assert.equal(completion.choices[0].message.content, "Browser");
+      assert.equal(completion.model, "glm-5.3");
      assert.equal(capturedRequest?.localStorage?.token, TEST_TOKEN);
      assert.equal(capturedRequest?.localStorageOrigin, "https://chat.z.ai");
+      assert.equal(capturedRequest?.headless, false);
      assert.equal(capturedRequest?.inputSelector, "#chat-input");
      assert.equal(
        capturedRequest?.submitButtonSelector,
@@ -331,7 +338,7 @@ describe("ZaiWebExecutor", () => {
      );
      assert.equal(capturedRequest?.submitButtonMode, "dom");
      assert.equal(capturedRequest?.userMessage, "hi");
-      assert.match(capturedRequest?.chatPageUrl ?? "", /model=GLM-5\.2/);
+      assert.match(capturedRequest?.chatPageUrl ?? "", /model=GLM-5\.3/);
      assert.equal(typeof capturedRequest?.beforeSubmit, "function");
      assert.equal(result.headers["X-OmniRoute-Transport"], "browser");
      assert.equal(result.transformedBody.browser_backed, true);
@@ -342,7 +349,7 @@ describe("ZaiWebExecutor", () => {
    }
  });

-  it("configures GLM-5V-Turbo controls on the browser transport", async () => {
+  it("configures GLM-5.3-Flash on the browser transport", async () => {
    let capturedRequest: BrowserBackedChatRequest | null = null;
    browserChat.__setBrowserBackedChatOverrideForTesting(async (request) => {
      capturedRequest = request;
@@ -352,8 +359,8 @@ describe("ZaiWebExecutor", () => {
    try {
      const executor = new mod.ZaiWebExecutor();
      const result = await executor.execute({
-        model: "GLM-5v-Turbo",
-        body: { messages: [{ role: "user", content: "use the model tools" }] },
+        model: "glm-5.3-flash",
+        body: { messages: [{ role: "user", content: "use flash" }] },
        stream: false,
        credentials: { apiKey: TEST_TOKEN },
        signal: null,
@@ -361,19 +368,19 @@ describe("ZaiWebExecutor", () => {

      const completion = await result.response.json();
      assert.equal(completion.choices[0].message.content, "VLM");
-      assert.match(capturedRequest?.chatPageUrl ?? "", /model=GLM-5V-Turbo/);
+      assert.match(capturedRequest?.chatPageUrl ?? "", /model=GLM-5\.3-Flash/);
      assert.equal(typeof capturedRequest?.beforeSubmit, "function");
      assert.equal(result.transformedBody.enable_thinking, true);
-      assert.equal(result.transformedBody.vlm_tools_enable, true);
-      assert.equal(result.transformedBody.vlm_web_search_enable, true);
-      assert.equal(result.transformedBody.vlm_website_mode, true);
-      assert.equal("reasoning_effort" in result.transformedBody, false);
+      assert.equal(result.transformedBody.reasoning_effort, "max");
+      assert.equal(result.transformedBody.vlm_tools_enable, false);
+      assert.equal(result.transformedBody.vlm_web_search_enable, false);
+      assert.equal(result.transformedBody.vlm_website_mode, false);
    } finally {
      browserChat.__resetBrowserBackedChatOverrideForTesting();
    }
  });

-  it("uploads GLM-5V-Turbo image input through the authenticated browser page", async () => {
+  it("uploads GLM-5.3-Flash image input through the authenticated browser page", async () => {
    let capturedRequest: BrowserBackedChatRequest | null = null;
    browserChat.__setBrowserBackedChatOverrideForTesting(async (request) => {
      capturedRequest = request;
@@ -383,7 +390,7 @@ describe("ZaiWebExecutor", () => {
    try {
      const executor = new mod.ZaiWebExecutor();
      const result = await executor.execute({
-        model: "GLM-5v-Turbo",
+        model: "glm-5.3-flash",
        body: {
          messages: [
            {
@@ -445,7 +452,7 @@ describe("ZaiWebExecutor", () => {

    assert.equal(result.response.status, 400);
    const parsed = await result.response.json();
-    assert.match(parsed.error.message, /use GLM-5V-Turbo/);
+    assert.match(parsed.error.message, /use GLM-5\.3-Flash/);
  });

  it("creates a chat, signs the v2 request, and forwards the CAPTCHA proof", async () => {
@@ -461,9 +468,9 @@ describe("ZaiWebExecutor", () => {
    try {
      const executor = new mod.ZaiWebExecutor();
      const result = await executor.execute({
-        model: "GLM-5.1",
+        model: "glm-5.3",
        body: {
-          model: "GLM-5.1",
+          model: "glm-5.3",
          messages: [{ role: "user", content: "hello" }],
          temperature: 0.4,
          web_search: true,
@@ -477,7 +484,7 @@ describe("ZaiWebExecutor", () => {
      const newChatHeaders = capture.newChatInit?.headers as Record<string, string>;
      assert.equal(newChatHeaders.Authorization, `Bearer ${TEST_TOKEN}`);
      const newChatBody = JSON.parse(String(capture.newChatInit?.body));
-      assert.deepEqual(newChatBody.chat.models, ["GLM-5.1"]);
+      assert.deepEqual(newChatBody.chat.models, ["glm-5.3"]);
      assert.equal(newChatBody.chat.history.currentId.length, 36);
      assert.equal(newChatBody.chat.enable_thinking, true);
      assert.equal(newChatBody.chat.auto_web_search, true);
@@ -494,11 +501,11 @@ describe("ZaiWebExecutor", () => {

      const headers = capture.completionInit?.headers as Record<string, string>;
      assert.equal(headers.Authorization, `Bearer ${TEST_TOKEN}`);
-      assert.equal(headers["X-FE-Version"], "prod-fe-1.1.79");
+      assert.equal(headers["X-FE-Version"], "prod-fe-1.1.92");
      assert.match(headers["X-Signature"], /^[a-f0-9]{64}$/);

      const parsedBody = JSON.parse(String(capture.completionInit?.body));
-      assert.equal(parsedBody.model, "GLM-5.1");
+      assert.equal(parsedBody.model, "glm-5.3");
      assert.equal(parsedBody.stream, true);
      assert.deepEqual(parsedBody.messages, [{ role: "user", content: "hello" }]);
      assert.equal(parsedBody.signature_prompt, "hello");
@@ -508,7 +515,7 @@ describe("ZaiWebExecutor", () => {
      assert.equal(parsedBody.features.web_search, false);
      assert.equal(parsedBody.features.auto_web_search, true);
      assert.equal(parsedBody.features.enable_thinking, true);
-      assert.equal("reasoning_effort" in parsedBody.features, false);
+      assert.equal(parsedBody.features.reasoning_effort, "max");
      assert.equal(result.headers.Authorization, "Bearer [REDACTED]");
      assert.equal(result.transformedBody.captcha_verify_param, "[REDACTED]");
    } finally {
@@ -516,7 +523,7 @@ describe("ZaiWebExecutor", () => {
    }
  });

-  it("sends GLM-5.2 Deep Think High through the direct request path", async () => {
+  it("sends GLM-5.3 Deep Think Low through the direct request path", async () => {
    const capture: ZaiFetchCapture = {};
    const originalFetch = installZaiFetch(
      () =>
@@ -529,36 +536,36 @@ describe("ZaiWebExecutor", () => {
    try {
      const executor = new mod.ZaiWebExecutor();
      await executor.execute({
-        model: "glm-5.2",
+        model: "glm-5.3",
        body: {
-          model: "glm-5.2",
+          model: "glm-5.3",
          messages: [{ role: "user", content: "think carefully" }],
-          reasoning_effort: "high",
+          reasoning_effort: "low",
        },
        stream: false,
        credentials: { apiKey: TEST_CREDENTIAL },
        signal: null,
      });

-      // #8014: completions must target the versioned v2 path. The query string
+      // The query string
      // carries the per-request signature payload, so match the endpoint prefix.
      assert.ok(
        String(capture.completionUrl).startsWith("https://chat.z.ai/api/v2/chat/completions?"),
-        `expected the v2 completions endpoint, got ${capture.completionUrl}`
+        `expected the current completions endpoint, got ${capture.completionUrl}`
      );
      const newChatBody = JSON.parse(String(capture.newChatInit?.body));
      assert.equal(newChatBody.chat.enable_thinking, true);
-      assert.equal(newChatBody.chat.reasoning_effort, "high");
+      assert.equal(newChatBody.chat.reasoning_effort, "low");

      const completionBody = JSON.parse(String(capture.completionInit?.body));
      assert.equal(completionBody.features.enable_thinking, true);
-      assert.equal(completionBody.features.reasoning_effort, "high");
+      assert.equal(completionBody.features.reasoning_effort, "low");
    } finally {
      globalThis.fetch = originalFetch;
    }
  });

-  it("sends GLM-5V-Turbo VLM tools and web-search flags through the direct path", async () => {
+  it("maps GLM-5.3-Flash to its opaque wire id on the direct path", async () => {
    const capture: ZaiFetchCapture = {};
    const originalFetch = installZaiFetch(
      () =>
@@ -571,10 +578,11 @@ describe("ZaiWebExecutor", () => {
    try {
      const executor = new mod.ZaiWebExecutor();
      await executor.execute({
-        model: "GLM-5v-Turbo",
+        model: "glm-5.3-flash",
        body: {
-          model: "GLM-5v-Turbo",
-          messages: [{ role: "user", content: "inspect this image" }],
+          model: "glm-5.3-flash",
+          messages: [{ role: "user", content: "answer quickly" }],
+          web_search: true,
        },
        stream: false,
        credentials: { apiKey: TEST_CREDENTIAL },
@@ -582,19 +590,26 @@ describe("ZaiWebExecutor", () => {
      });

      const newChatBody = JSON.parse(String(capture.newChatInit?.body));
+      assert.deepEqual(newChatBody.chat.models, ["x-preview-l"]);
+      assert.deepEqual(
+        newChatBody.chat.history.messages[newChatBody.chat.history.currentId].models,
+        ["x-preview-l"]
+      );
      assert.equal(newChatBody.chat.enable_thinking, true);
      assert.equal(newChatBody.chat.auto_web_search, true);
-      assert.equal(newChatBody.chat.extra.vlm_tools_enable, true);
-      assert.equal(newChatBody.chat.extra.vlm_web_search_enable, true);
-      assert.equal(newChatBody.chat.extra.vlm_website_mode, true);
+      assert.equal(newChatBody.chat.reasoning_effort, "max");
+      assert.equal(newChatBody.chat.extra.vlm_tools_enable, false);
+      assert.equal(newChatBody.chat.extra.vlm_web_search_enable, false);
+      assert.equal(newChatBody.chat.extra.vlm_website_mode, false);

      const completionBody = JSON.parse(String(capture.completionInit?.body));
+      assert.equal(completionBody.model, "x-preview-l");
      assert.equal(completionBody.features.enable_thinking, true);
-      assert.equal(completionBody.features.auto_web_search, false);
-      assert.equal(completionBody.features.vlm_tools_enable, true);
-      assert.equal(completionBody.features.vlm_web_search_enable, true);
-      assert.equal(completionBody.features.vlm_website_mode, true);
-      assert.equal("reasoning_effort" in completionBody.features, false);
+      assert.equal(completionBody.features.reasoning_effort, "max");
+      assert.equal(completionBody.features.auto_web_search, true);
+      assert.equal(completionBody.features.vlm_tools_enable, false);
+      assert.equal(completionBody.features.vlm_web_search_enable, false);
+      assert.equal(completionBody.features.vlm_website_mode, false);
    } finally {
      globalThis.fetch = originalFetch;
    }
@@ -619,7 +634,7 @@ describe("ZaiWebExecutor", () => {
    try {
      const executor = new mod.ZaiWebExecutor();
      const result = await executor.execute({
-        model: "GLM-5.1",
+        model: "glm-5.3",
        body: { messages: [{ role: "user", content: "hi" }] },
        stream: false,
        credentials: { apiKey: TEST_CREDENTIAL },
@@ -651,7 +666,7 @@ describe("ZaiWebExecutor", () => {
    try {
      const executor = new mod.ZaiWebExecutor();
      const result = await executor.execute({
-        model: "GLM-5.1",
+        model: "glm-5.3",
        body: { messages: [{ role: "user", content: "hi" }] },
        stream: true,
        credentials: { apiKey: TEST_CREDENTIAL },
@@ -673,7 +688,7 @@ describe("ZaiWebExecutor", () => {
    try {
      const executor = new mod.ZaiWebExecutor();
      const result = await executor.execute({
-        model: "GLM-5.1",
+        model: "glm-5.3",
        body: { messages: [{ role: "user", content: "hi" }] },
        stream: false,
        credentials: { apiKey: TEST_CREDENTIAL },
--- a/tests/unit/model-test-runner.test.ts
+++ b/tests/unit/model-test-runner.test.ts
@@ -314,7 +314,7 @@ test("resolveModelTestTimeoutMs defaults ordinary model checks to 30 seconds", (

 test("resolveModelTestTimeoutMs gives zai-web checks up to 60 seconds", () => {
  assert.equal(resolveModelTestTimeoutMs("zai-web", "glm-5.2", 30_000), 60_000);
-  assert.equal(resolveModelTestTimeoutMs("zai-web", "zai-web/GLM-5V-Turbo", 90_000), 90_000);
+  assert.equal(resolveModelTestTimeoutMs("zai-web", "zai-web/glm-5.3-flash", 90_000), 90_000);
 });

 // ---------------------------------------------------------------------------
--- a/tests/unit/zai-web-chat-endpoint-8014-probe.test.ts
+++ b/tests/unit/zai-web-chat-endpoint-8014-probe.test.ts
@@ -9,7 +9,7 @@ const TEST_TOKEN = "e30.eyJpZCI6InVzZXItMTIzIn0.sig";
 /**
 * #8014 guard: the executor must target the versioned v2 completions endpoint,
 * never the stale unversioned `/api/chat/completions` path, which 404s
- * model-independently as of 2026-07.
+ * model-independently.
 *
 * Setup notes for this flow (the executor now creates a remote chat first and
 * signs the completion request):
@@ -43,7 +43,7 @@ test("#8014: ZaiWebExecutor must POST to the current chat.z.ai v2 chat-completio
  try {
    const executor = new mod.ZaiWebExecutor();
    const result = await executor.execute({
-      model: "glm-4.6",
+      model: "glm-5.3",
      body: { messages: [{ role: "user", content: "hello" }] },
      stream: false,
      credentials: {
@@ -55,7 +55,6 @@ test("#8014: ZaiWebExecutor must POST to the current chat.z.ai v2 chat-completio
    assert.ok(requested.length > 0, "the direct path must actually reach fetch");

    assert.ok(
-      // Exact-URL match (not a substring test): `requested` holds whole URLs.
      !requested.some((url) => url === STALE_URL),
      `zai-web executor POSTed to the stale endpoint — matches #8014's model-independent 404 "Not Found"`
    );
--- a/tests/unit/zai-web-models-discovery-7678.test.ts
+++ b/tests/unit/zai-web-models-discovery-7678.test.ts
@@ -13,7 +13,7 @@ const providersDb = await import("../../src/lib/db/providers.ts");
 const modelsRoute = await import("../../src/app/api/providers/[id]/models/route.ts");
 const registry = await import("../../open-sse/config/providers/registry/zai-web/index.ts");

-const CURATED_ZAI_WEB_MODEL_IDS = ["glm-5.2", "GLM-5.1", "GLM-5-Turbo", "GLM-5v-Turbo"];
+const CURATED_ZAI_WEB_MODEL_IDS = ["glm-5.3-flash", "glm-5.3", "glm-5.2"];

 async function resetStorage() {
  core.resetDbInstance();
@@ -31,34 +31,32 @@ test("zai-web publishes the live reasoning and vision capabilities", () => {
    registry.zai_webProvider.models.map((model) => ({
      id: model.id,
      supportsReasoning: model.supportsReasoning === true,
+      supportedThinkingEfforts: model.supportedThinkingEfforts,
      supportsVision: model.supportsVision === true,
      toolCalling: model.toolCalling === true,
    })),
    [
+      {
+        id: "glm-5.3-flash",
+        supportsReasoning: true,
+        supportedThinkingEfforts: ["low", "high", "max"],
+        supportsVision: true,
+        toolCalling: false,
+      },
+      {
+        id: "glm-5.3",
+        supportsReasoning: true,
+        supportedThinkingEfforts: ["low", "high", "max"],
+        supportsVision: false,
+        toolCalling: false,
+      },
      {
        id: "glm-5.2",
        supportsReasoning: true,
+        supportedThinkingEfforts: ["high", "max"],
        supportsVision: false,
        toolCalling: false,
      },
-      {
-        id: "GLM-5.1",
-        supportsReasoning: true,
-        supportsVision: false,
-        toolCalling: false,
-      },
-      {
-        id: "GLM-5-Turbo",
-        supportsReasoning: true,
-        supportsVision: false,
-        toolCalling: false,
-      },
-      {
-        id: "GLM-5v-Turbo",
-        supportsReasoning: true,
-        supportsVision: true,
-        toolCalling: false,
-      },
    ]
  );
 });