@bitkyc08/opencodex 2.29.0 → 2.31.0-preview.20260822

package/src/providers/cursor-pool.ts

@@ -0,0 +1,72 @@
+/**
+ * Weighted credential routing for Cursor accounts.
+ *
+ * Transfer from yelixir-dev/cursor-ai-proxy-bridge credentials.ts:
+ * weighted round-robin selection with per-credential auth-failure cooldown
+ * and one-retry failover on a different account before surfacing the error.
+ *
+ * OpenCodex already has JWT-based multi-account identification (src/oauth/cursor.ts)
+ * and Anthropic-specific 429 rotation; this module adds Cursor-aware weighted
+ * routing on top of those primitives.
+ */
+
+export interface CursorCredential {
+  readonly id: string;
+  weight: number;
+}
+
+interface CredentialState {
+  readonly credential: CursorCredential;
+  currentWeight: number;
+  disabledUntil: number;
+}
+
+export class NoAvailableCursorCredentialError extends Error {
+  constructor(message = "No available Cursor credentials") { super(message); }
+}
+
+export class CursorCredentialRouter {
+  private states: CredentialState[] = [];
+  private readonly cooldownMs: number;
+
+  constructor(credentials: ReadonlyArray<CursorCredential>, cooldownMs = 300_000) {
+    this.cooldownMs = cooldownMs;
+    this.replace(credentials);
+  }
+
+  replace(credentials: ReadonlyArray<CursorCredential>): void {
+    this.states = credentials.map(c => ({
+      credential: { ...c, weight: Math.max(1, c.weight || 1) },
+      currentWeight: 0,
+      disabledUntil: 0,
+    }));
+  }
+
+  pick(excludeIds: ReadonlySet<string> = new Set()): CursorCredential {
+    const now = Date.now();
+    const candidates = this.states.filter(s =>
+      !excludeIds.has(s.credential.id) && s.disabledUntil <= now,
+    );
+    if (candidates.length === 0) throw new NoAvailableCursorCredentialError();
+    let selected: CredentialState | undefined;
+    let totalWeight = 0;
+    for (const state of candidates) {
+      state.currentWeight += state.credential.weight;
+      totalWeight += state.credential.weight;
+      if (!selected || state.currentWeight > selected.currentWeight) selected = state;
+    }
+    if (!selected) throw new NoAvailableCursorCredentialError();
+    selected.currentWeight -= totalWeight;
+    return { ...selected.credential };
+  }
+
+  disable(id: string): void {
+    const state = this.states.find(s => s.credential.id === id);
+    if (state) state.disabledUntil = Date.now() + this.cooldownMs;
+  }
+
+  get snapshot(): ReadonlyArray<{ id: string; disabled: boolean }> {
+    const now = Date.now();
+    return this.states.map(s => ({ id: s.credential.id, disabled: s.disabledUntil > now }));
+  }
+}

package/src/providers/derive.ts

@@ -483,6 +483,9 @@ export function enrichProviderFromRegistry(name: string, prov: OcxProviderConfig
   if (prov.supportsOpenAiWebSearchToolFields === undefined && entry.supportsOpenAiWebSearchToolFields !== undefined) {
     prov.supportsOpenAiWebSearchToolFields = entry.supportsOpenAiWebSearchToolFields;
   }
+  if (prov.supportsResponsesCustomTools === undefined && entry.supportsResponsesCustomTools !== undefined) {
+    prov.supportsResponsesCustomTools = entry.supportsResponsesCustomTools;
+  }
   if (prov.preserveResponsesReasoningContent === undefined && entry.preserveResponsesReasoningContent !== undefined) prov.preserveResponsesReasoningContent = entry.preserveResponsesReasoningContent;
   applyReasoningSummaryDefaults(prov, entry.modelSupportsReasoningSummaries);
   applyServiceTierModelDefaults(prov, serviceTierModelDefaultsFor(entry, prov));

package/src/providers/fastwire.ts

@@ -33,6 +33,7 @@ export interface FastPolicyAuthority {
   readonly providerAdapter: string;
   readonly providerAuthMode?: ProviderAuthKind;
   readonly fastWireDeclaration: FastWire | null | undefined;
+  readonly fastTierDescription?: string;
   readonly modelWireOverrideAllowed: boolean;
   readonly authTransport: FastPolicyAuthTransport;
   readonly capability: {
@@ -55,6 +56,7 @@ export interface ResolvedFastPolicy {
     | "pin-unavailable";
   readonly adapter: string;
   readonly fastWire: FastWire | null;
+  readonly fastTierDescription?: string;
   readonly forwardCallerTier: boolean;
 }
 
@@ -222,7 +224,16 @@ export function resolveFastPolicy(
   else if (capability === undefined) eligibility = "unclassified";
   else eligibility = "eligible";
 
-  return { capability, eligibility, adapter, fastWire, forwardCallerTier };
+  return {
+    capability,
+    eligibility,
+    adapter,
+    fastWire,
+    ...(authority.fastTierDescription !== undefined
+      ? { fastTierDescription: authority.fastTierDescription }
+      : {}),
+    forwardCallerTier,
+  };
 }
 
 export function canonicalFastTierMarker(callerTier: string | undefined): "priority" | undefined {

package/src/providers/openai-sidecar.ts

@@ -169,6 +169,7 @@ export async function resolveFirstUsableOpenAiSidecar(
             authContext.accountId,
             outcome,
             {
+              threadId: authContext.affinityKey,
               probeLeaseId: authContext.probeLeaseId,
               writerGeneration: authContext.writerGeneration,
             },

package/src/providers/quota.ts

@@ -50,6 +50,7 @@ const OPENROUTER_BASE_URL = "https://openrouter.ai/api/v1";
 const DEEPSEEK_BASE_URL = "https://api.deepseek.com";
 const CLINE_BASE_URL = "https://api.cline.bot";
 const ZAI_BASE_URL = "https://api.z.ai";
+const ZAI_CN_BASE_URL = "https://open.bigmodel.cn";
 const MINIMAX_REMAINS_URL = "https://www.minimax.io/v1/token_plan/remains";
 const MOONSHOT_BASE_URL = "https://api.moonshot.ai/v1";
 const VENICE_BASE_URL = "https://api.venice.ai/api/v1";
@@ -343,7 +344,12 @@ function isCanonicalClineBaseUrl(baseUrl: string): boolean {
 
 function isCanonicalZaiBaseUrl(baseUrl: string): boolean {
   const normalized = normalizedBaseUrl(baseUrl);
-  return normalized === ZAI_BASE_URL || normalized === `${ZAI_BASE_URL}/api/coding/paas/v4`;
+  return normalized === ZAI_BASE_URL
+    || normalized === `${ZAI_BASE_URL}/api/coding/paas/v4`
+    || normalized === ZAI_CN_BASE_URL
+    || normalized === `${ZAI_CN_BASE_URL}/api/coding/paas/v4`
+    // BigModel serves the same GLM Coding Plan on the OpenAI Responses wire at /api/v1.
+    || normalized === `${ZAI_CN_BASE_URL}/api/v1`;
 }
 
 function isCanonicalMinimaxBaseUrl(baseUrl: string): boolean {
@@ -669,34 +675,67 @@ async function fetchClineQuota(provider: string, config: OcxProviderConfig): Pro
 
 /**
  * Z.AI GLM Coding Plan `GET /api/monitor/usage/quota/limit` — the coding-plan
- * subscription's 5-hour token cycle, weekly quota, and monthly MCP usage.
- * Authenticates with the API key as a Bearer token per Z.AI's API reference.
+ * limits arrive as a `limits` array of `TOKENS_LIMIT` (newer plans call the
+ * same rows `CREDIT_LIMIT`) and `TIME_LIMIT` rows. `TOKENS_LIMIT`/`CREDIT_LIMIT`
+ * rows carry the window length as `unit`/`number`: unit 3 is hours (number 5 →
+ * the rolling five-hour window), unit 6 is weeks (number 1 → the weekly
+ * window). `TIME_LIMIT` rows are the monthly MCP tool budget (Web Search / Web
+ * Reader / Zread). Every row's `percentage` is the consumed share (falling
+ * back to `currentValue`/`usage` when absent) and `nextResetTime` (unix ms)
+ * the window reset.
  */
-async function fetchZaiQuota(provider: string, config: OcxProviderConfig): Promise<ProviderQuotaProbeResult> {
-  if (!isCanonicalZaiBaseUrl(config.baseUrl)) return null;
-  const apiKey = resolveEnvValue(config.apiKey)?.trim();
-  if (!apiKey) return null;
-  const response = await fetch(`${ZAI_BASE_URL}/api/monitor/usage/quota/limit`, {
-    headers: { Accept: "application/json", Authorization: `Bearer ${apiKey}` },
-    redirect: "error",
-    signal: AbortSignal.timeout(REQUEST_TIMEOUT_MS),
-  });
-  if (!response.ok) {
-    return response.status >= 400 && response.status < 500 && response.status !== 408 && response.status !== 429
-      ? TERMINAL_QUOTA_FAILURE
-      : null;
+export function parseZaiQuotaLimits(data: Record<string, unknown> | null): ProviderQuota | null {
+  const limits = Array.isArray(data?.limits) ? data.limits as unknown[] : null;
+  if (!limits) return null;
+  const quota: ProviderQuota = { updatedAt: Date.now() };
+  let windows = 0;
+  for (const raw of limits) {
+    const row = asRecord(raw);
+    if (!row) continue;
+    const resetAt = normalizeResetAt(row.nextResetTime);
+    let percent = normalizePercent(row.percentage);
+    if (percent === undefined) {
+      const used = toFiniteNumber(row.currentValue);
+      const total = toFiniteNumber(row.usage);
+      if (used !== undefined && total !== undefined && total > 0) {
+        percent = normalizePercent((used / total) * 100);
+      }
+    }
+    if (percent === undefined) continue;
+    if (row.type === "TOKENS_LIMIT" || row.type === "CREDIT_LIMIT") {
+      const unit = toFiniteNumber(row.unit);
+      const number = toFiniteNumber(row.number);
+      if (unit === 3 && number === 5) {
+        quota.fiveHourPercent = percent;
+        if (resetAt !== undefined) quota.fiveHourResetAt = resetAt;
+        windows += 1;
+      } else if (unit === 6 && number === 1) {
+        quota.weeklyPercent = percent;
+        if (resetAt !== undefined) quota.weeklyResetAt = resetAt;
+        windows += 1;
+      }
+    } else if (row.type === "TIME_LIMIT") {
+      quota.monthlyPercent = percent;
+      if (resetAt !== undefined) quota.monthlyResetAt = resetAt;
+      windows += 1;
+    }
   }
-  const body = asRecord(await readQuotaJson(response));
-  if (!body || body.success === false) return null;
-  const data = asRecord(body.data) ?? body;
-  // The plugin renders a 5h token window, a weekly window, and a monthly MCP
-  // window. Look for percent fields with window identifiers.
+  return windows > 0 ? quota : null;
+}
+
+/**
+ * Legacy Z.AI payload shape: percent fields with window identifiers directly on
+ * the data object (optionally nested under `quota`). Kept as a fallback so
+ * older responses keep rendering when the `limits` array is absent.
+ */
+function parseZaiQuotaLegacyFields(data: Record<string, unknown> | null): ProviderQuota | null {
+  if (!data) return null;
   const quota: ProviderQuota = { updatedAt: Date.now() };
   let windows = 0;
   const percentAt = (key: string): number | undefined => {
-    const value = normalizePercent(data?.[key]);
+    const value = normalizePercent(data[key]);
     if (value !== undefined) return value;
-    const nested = asRecord(data?.quota);
+    const nested = asRecord(data.quota);
     return nested ? normalizePercent(nested[key]) : undefined;
   };
   const fiveHour = percentAt("fiveHourPercent") ?? percentAt("fiveHourUsage") ?? percentAt("fiveHourUsed");
@@ -714,7 +753,40 @@ async function fetchZaiQuota(provider: string, config: OcxProviderConfig): Promi
     quota.monthlyPercent = monthly;
     windows += 1;
   }
-  return windows > 0 ? report(provider, "zai:quota-limit", quota) : null;
+  return windows > 0 ? quota : null;
+}
+
+/**
+ * Fetches the Z.AI GLM Coding Plan quota — on whichever region the provider
+ * points at (api.z.ai or open.bigmodel.cn). Authenticates with the API key as
+ * a Bearer token per Z.AI's API reference. The `limits` array shape is
+ * preferred; older field-name payloads fall back to the legacy parser.
+ */
+async function fetchZaiQuota(provider: string, config: OcxProviderConfig): Promise<ProviderQuotaProbeResult> {
+  if (!isCanonicalZaiBaseUrl(config.baseUrl)) return null;
+  const apiKey = resolveEnvValue(config.apiKey)?.trim();
+  if (!apiKey) return null;
+  const normalized = normalizedBaseUrl(config.baseUrl);
+  const monitorHost = normalized === ZAI_BASE_URL || normalized === `${ZAI_BASE_URL}/api/coding/paas/v4`
+    ? ZAI_BASE_URL
+    : ZAI_CN_BASE_URL;
+  const response = await fetch(`${monitorHost}/api/monitor/usage/quota/limit`, {
+    headers: { Accept: "application/json", Authorization: `Bearer ${apiKey}` },
+    redirect: "error",
+    signal: AbortSignal.timeout(REQUEST_TIMEOUT_MS),
+  });
+  if (!response.ok) {
+    return response.status >= 400 && response.status < 500 && response.status !== 408 && response.status !== 429
+      ? TERMINAL_QUOTA_FAILURE
+      : null;
+  }
+  const body = asRecord(await readQuotaJson(response));
+  if (!body || body.success === false) return null;
+  const data = asRecord(body.data) ?? body;
+  const quota = Array.isArray(data?.limits)
+    ? parseZaiQuotaLimits(data)
+    : parseZaiQuotaLegacyFields(data);
+  return quota ? report(provider, "zai:quota-limit", quota) : null;
 }
 
 /**
@@ -2106,7 +2178,8 @@ async function maybeFetchProviderQuota(
     if ((provider.authMode ?? "key") === "key" && name === "cline-pass") {
       return fetchClineQuota(name, provider);
     }
-    if ((provider.authMode ?? "key") === "key" && name === "zai") {
+    if ((provider.authMode ?? "key") === "key"
+      && (name === "zai" || name === "glm" || name === "glm-cn" || name === "zhipu-bigmodel-coding")) {
       return fetchZaiQuota(name, provider);
     }
     if ((provider.authMode ?? "key") === "key" && (name === "minimax" || name === "minimax-cn")) {

package/src/providers/registry.ts

@@ -10,6 +10,7 @@ import {
   MOONSHOT_BASE_URL_CHOICES, MOONSHOT_INTL_BASE_URL,
 } from "./base-url-choices";
 import {
+  CURSOR_NO_VISION_MODELS,
   CURSOR_STATIC_MODELS,
   cursorModelContextWindows,
   cursorModelIds,
@@ -224,8 +225,22 @@ export interface ProviderRegistryEntry {
   supportsServiceTier?: boolean;
   /** Registry default for OpenAI extended hosted web_search field support. */
   supportsOpenAiWebSearchToolFields?: boolean;
+  /** Registry default for native Responses custom-tool support. */
+  supportsResponsesCustomTools?: boolean;
   /** Registry default for exact model service-tier capability; explicit config keys win. */
   modelSupportsServiceTier?: Record<string, boolean>;
+  /**
+   * Registry-only service-tier defaults for an OAuth preset's explicit API-key transport.
+   * Applied only when `allowKeyAuthOverride` is true and the captured effective auth transport
+   * is key-based. Explicit provider config still wins field-by-field, including `false`.
+   */
+  keyAuthServiceTier?: {
+    supportsServiceTier?: boolean;
+    modelSupportsServiceTier?: Record<string, boolean>;
+    chatServiceTier?: boolean;
+  };
+  /** Provider-specific copy for the Codex catalog's Fast tier. */
+  fastTierDescription?: string;
   /**
    * Registry-only destination guard for `modelSupportsServiceTier`. This scopes vendor evidence
    * without changing provider ownership, routing, authentication, or config validation.
@@ -452,7 +467,23 @@ const THINKING_BUDGET_MODELS = [
 ];
 const OPENCODE_GO_THINKING_BUDGET_MODELS = ["qwen3.5-plus", "qwen3.6-plus", "qwen3.7-max", "qwen3.7-plus"];
 const DEEPSEEK_THINKING_MODELS = ["deepseek-v4-pro", "deepseek-v4-flash"];
+/*
+ * DeepSeek's experimental vision preview (released 2026-08-21, api-docs.deepseek.com):
+ * text+image input on the V4 Flash base. DeepSeek positions it as a preview id;
+ * the expectation is that vision merges into `deepseek-v4-flash` proper later,
+ * at which point this id retires the same way deepseek-chat/reasoner did.
+ */
+const DEEPSEEK_VISION_PREVIEW_MODEL = "deepseek-v4-flash-vision-exp";
 const OPENCODE_FREE_DEEPSEEK_MODELS = ["deepseek-v4-flash-free"];
+/*
+ * OpenCode Zen's free slug for the OpenRouter stealth model "Ox Alpha"
+ * (openrouter.ai/stealth/ox-alpha): 1,048,576-token context, multimodal
+ * (text+image+video upstream; Zen serves text+image), mandatory reasoning,
+ * free during the stealth window. Zen displays it as "Ox Alpha Free" under
+ * this exact id (opencode.ai/docs/zen, verified 2026-08-21).
+ */
+const OPENCODE_OX_ALPHA_FREE_MODEL = "x-preview-f-free";
+const OX_ALPHA_CONTEXT_WINDOW = 1_048_576;
 /*
  * Zen free models that reject `image_url` upstream (#1043, and the reproducible
  * half of #1024).
@@ -994,11 +1025,10 @@ export const PROVIDER_REGISTRY: readonly ProviderRegistryEntry[] = [
     // no-effort fallback to `kimi-k3-max` would never be reached. Mirrors the other K3
     // routes (kimi, kimi-code, opencode-go).
     modelDefaultReasoningEfforts: { "kimi-k3": "max" },
-    // Cursor's wire protocol never forwards image parts (request-builder emits an unsupported-
-    // content marker), so the vision sidecar covers ALL cursor models regardless of what the
-    // upstream model could natively do. Live-discovered models outside the static list fall back
-    // to the same marker until they appear here.
-    noVisionModels: cursorModelIds(CURSOR_STATIC_MODELS),
+    // Blind Cursor models (Auto routers, Composer, GLM-5.2, GLM-5.3) go through the vision sidecar;
+    // multimodal hosts (Claude/Gemini/GPT/Kimi/Grok) take native SelectedImage. The catalog
+    // still advertises image for noVision members so Codex can attach (sidecar option B).
+    noVisionModels: [...CURSOR_NO_VISION_MODELS],
   },
   {
     id: "xai",
@@ -1007,10 +1037,21 @@ export const PROVIDER_REGISTRY: readonly ProviderRegistryEntry[] = [
     baseUrl: "https://api.x.ai/v1",
     authKind: "oauth",
     allowKeyAuthOverride: true,
+    // Priority Processing is documented for xAI's public API-key Chat Completions and
+    // Responses endpoints. OAuth is a separate Grok CLI subscription gateway and remains
+    // unclassified; do not turn this into a provider-wide supportsServiceTier declaration.
+    keyAuthServiceTier: {
+      supportsServiceTier: true,
+      chatServiceTier: true,
+    },
+    fastTierDescription: "Priority processing, 2x token price",
     featured: true,
     oauthId: "xai",
     jawcodeBundle: "xai",
     supportsOpenAiWebSearchToolFields: false,
+    // Live A/B on 2026-08-20: xAI rejects native custom/custom_tool_call shapes while accepting
+    // the otherwise-identical request after the custom tool is lowered to a function.
+    supportsResponsesCustomTools: false,
     note: "Log in with your Grok account",
     // Parallel tool calls: officially supported and default-on per docs.x.ai function-calling
     // (verified 260709, devlog/_plan/260709_parallel_tool_calls). Streamed calls arrive whole
@@ -1101,6 +1142,17 @@ export const PROVIDER_REGISTRY: readonly ProviderRegistryEntry[] = [
     // Unknown/new live models deliberately do not advertise a reasoning picker.
     reasoningEfforts: [],
     modelReasoningEfforts: COMMAND_CODE_MODEL_REASONING_EFFORTS,
+    // Ox Alpha (stealth preview, changelog v1.31.0): free 1M multimodal reasoning
+    // model on every plan. DeepSeek vision preview id is preemptive metadata —
+    // it is expected to merge into deepseek-v4-flash later.
+    modelContextWindows: {
+      "stealth/ox-alpha": OX_ALPHA_CONTEXT_WINDOW,
+      [`deepseek/${DEEPSEEK_VISION_PREVIEW_MODEL}`]: 1_048_576,
+    },
+    modelInputModalities: {
+      "stealth/ox-alpha": ["text", "image"],
+      [`deepseek/${DEEPSEEK_VISION_PREVIEW_MODEL}`]: ["text", "image"],
+    },
     defaultMaxOutputTokens: 64_000,
     // The proprietary generate wire has no verified per-request serialization flag.
     parallelToolCalls: false,
@@ -1292,8 +1344,20 @@ export const PROVIDER_REGISTRY: readonly ProviderRegistryEntry[] = [
     - 장점, 단점 및 영향: Luna reaches `/responses` from every inbound surface without changing siblings; a future upstream endpoint change requires an evidence-backed registry update.
     */
     modelWireDefaults: { "gpt-5.6-luna": "openai-responses" },
-    modelContextWindows: { "kimi-k3": KIMI_K3_STANDARD_CONTEXT_WINDOW },
-    modelInputModalities: { "kimi-k3": ["text", "image"] },
+    modelContextWindows: {
+      "kimi-k3": KIMI_K3_STANDARD_CONTEXT_WINDOW,
+      // Ox Alpha (stealth 1M multimodal) and the DeepSeek vision preview are
+      // metadata-only here: the Go roster is discovered live, so these apply
+      // the moment the gateway starts serving the ids.
+      [OPENCODE_OX_ALPHA_FREE_MODEL]: OX_ALPHA_CONTEXT_WINDOW,
+      [DEEPSEEK_VISION_PREVIEW_MODEL]: 1_048_576,
+    },
+    modelInputModalities: {
+      "kimi-k3": ["text", "image"],
+      [OPENCODE_OX_ALPHA_FREE_MODEL]: ["text", "image"],
+      // Experimental DeepSeek vision preview — expected to merge into deepseek-v4-flash later.
+      [DEEPSEEK_VISION_PREVIEW_MODEL]: ["text", "image"],
+    },
     modelReasoningEfforts: {
       "glm-5.3": ZAI_GLM_53_REASONING_EFFORTS,
       "glm-5.2": ZAI_GLM_52_REASONING_EFFORTS,
@@ -1396,11 +1460,16 @@ export const PROVIDER_REGISTRY: readonly ProviderRegistryEntry[] = [
     featured: true,
     dashboardUrl: "https://openrouter.ai/keys",
     jawcodeBundle: "openrouter",
-    models: ["anthropic/claude-sonnet-5", ...OPENROUTER_GPT56_MODELS],
+    // stealth/ox-alpha: free stealth-window frontier model (launched 2026-08-20).
+    // /api/v1/models reports 1,048,576 context, 131,072 max output, text+image+video
+    // input, $0 pricing, mandatory reasoning. Single provider slug: `stealth`.
+    models: ["anthropic/claude-sonnet-5", "stealth/ox-alpha", ...OPENROUTER_GPT56_MODELS],
     modelContextWindows: {
       "anthropic/claude-sonnet-5": 1_000_000,
+      "stealth/ox-alpha": OX_ALPHA_CONTEXT_WINDOW,
       ...OPENROUTER_GPT56_CONTEXT_WINDOWS,
     },
+    modelInputModalities: { "stealth/ox-alpha": ["text", "image"] },
     // OpenRouter documents priority support for OpenAI endpoints, but not Anthropic. Keep the
     // provider unclassified and opt in only the exact OpenAI-backed slugs we ship. These facts
     // belong only to the canonical destination; a same-named custom gateway is unknown to us.
@@ -1551,11 +1620,14 @@ export const PROVIDER_REGISTRY: readonly ProviderRegistryEntry[] = [
     // keep validating and routing (they previously mapped to v4-flash; devlog
     // _fin/260710_provider_hardening/002_research_cn.md). The current offerings are
     // the V4 ids — defaultModel and the model-specific wiring above use them.
-    models: ["deepseek-chat", "deepseek-reasoner", ...DEEPSEEK_THINKING_MODELS],
+    // deepseek-v4-flash-vision-exp: experimental vision preview (2026-08-21) —
+    // expected to merge into deepseek-v4-flash later; see DEEPSEEK_VISION_PREVIEW_MODEL.
+    models: ["deepseek-chat", "deepseek-reasoner", ...DEEPSEEK_THINKING_MODELS, DEEPSEEK_VISION_PREVIEW_MODEL],
     defaultModel: "deepseek-v4-flash",
     // Official DeepSeek Codex setup (codex-deepseek-setup.sh) advertises 1,048,576
     // for both V4 models; the older 1,000,000 figure was a rounded approximation.
-    modelContextWindows: { "deepseek-v4-flash": 1_048_576, "deepseek-v4-pro": 1_048_576 },
+    modelContextWindows: { "deepseek-v4-flash": 1_048_576, "deepseek-v4-pro": 1_048_576, [DEEPSEEK_VISION_PREVIEW_MODEL]: 1_048_576 },
+    modelInputModalities: { [DEEPSEEK_VISION_PREVIEW_MODEL]: ["text", "image"] },
     // DeepSeek documents both V4 models as native Responses API models adapted for Codex
     // (model table marks Responses API ✓ for flash and pro; the /responses reference lists
     // both ids as accepted `model` values — verified 2026-08-13 with the V4 Pro GA,
@@ -1808,6 +1880,17 @@ export const PROVIDER_REGISTRY: readonly ProviderRegistryEntry[] = [
     // slash ids — so a Codex-facing slug like `commandcode/deepseek-deepseek-v4-pro`
     // is sent upstream verbatim and rejected with `unsupported_model`.
     modelReasoningEfforts: COMMAND_CODE_MODEL_REASONING_EFFORTS,
+    // Ox Alpha (stealth preview, Command Code changelog v1.31.0) ships with a
+    // 1.05M-token multimodal context; the DeepSeek vision preview id is
+    // preemptive for when the catalog serves it (merges into v4-flash later).
+    modelContextWindows: {
+      "stealth/ox-alpha": OX_ALPHA_CONTEXT_WINDOW,
+      [`deepseek/${DEEPSEEK_VISION_PREVIEW_MODEL}`]: 1_048_576,
+    },
+    modelInputModalities: {
+      "stealth/ox-alpha": ["text", "image"],
+      [`deepseek/${DEEPSEEK_VISION_PREVIEW_MODEL}`]: ["text", "image"],
+    },
     modelDiscovery: {
       path: "models",
       maxResponseBytes: 256 * 1024,
@@ -2452,6 +2535,16 @@ export const PROVIDER_REGISTRY: readonly ProviderRegistryEntry[] = [
       [...DEEPSEEK_THINKING_MODELS, ...OPENCODE_FREE_DEEPSEEK_MODELS].map(id => [id, deepseekReasoningMapFor(id)]),
     ),
     preserveReasoningContentModels: [...DEEPSEEK_THINKING_MODELS, ...OPENCODE_FREE_DEEPSEEK_MODELS],
+    // Same Zen gateway as opencode-free: Ox Alpha Free (1M multimodal stealth model)
+    // and the DeepSeek vision preview (merges into deepseek-v4-flash later).
+    modelContextWindows: {
+      [OPENCODE_OX_ALPHA_FREE_MODEL]: OX_ALPHA_CONTEXT_WINDOW,
+      [DEEPSEEK_VISION_PREVIEW_MODEL]: 1_048_576,
+    },
+    modelInputModalities: {
+      [OPENCODE_OX_ALPHA_FREE_MODEL]: ["text", "image"],
+      [DEEPSEEK_VISION_PREVIEW_MODEL]: ["text", "image"],
+    },
     noVisionModels: [...OPENCODE_ZEN_TEXT_ONLY_MODELS, ...DEEPSEEK_THINKING_MODELS],
   },
   { id: "vercel-ai-gateway", label: "Vercel AI Gateway", baseUrl: "https://ai-gateway.vercel.sh/v1", adapter: "openai-chat", authKind: "key", dashboardUrl: "https://vercel.com/dashboard" },
@@ -2482,6 +2575,18 @@ export const PROVIDER_REGISTRY: readonly ProviderRegistryEntry[] = [
     modelReasoningEfforts: Object.fromEntries(OPENCODE_FREE_DEEPSEEK_MODELS.map(id => [id, deepseekThinkingEffortsFor(id)])),
     modelReasoningEffortMap: Object.fromEntries(OPENCODE_FREE_DEEPSEEK_MODELS.map(id => [id, deepseekReasoningMapFor(id)])),
     preserveReasoningContentModels: OPENCODE_FREE_DEEPSEEK_MODELS,
+    // Ox Alpha Free (`x-preview-f-free`): the OpenRouter stealth model on Zen's
+    // free tier — 1,048,576 context, text+image input. Deliberately NOT in the
+    // text-only list below. The DeepSeek vision preview id is preemptive
+    // metadata for when Zen starts serving it (merges into v4-flash later).
+    modelContextWindows: {
+      [OPENCODE_OX_ALPHA_FREE_MODEL]: OX_ALPHA_CONTEXT_WINDOW,
+      [DEEPSEEK_VISION_PREVIEW_MODEL]: 1_048_576,
+    },
+    modelInputModalities: {
+      [OPENCODE_OX_ALPHA_FREE_MODEL]: ["text", "image"],
+      [DEEPSEEK_VISION_PREVIEW_MODEL]: ["text", "image"],
+    },
     // Same Zen roster behind the same base URL, so it carries the same measured
     // text-only list rather than only its DeepSeek member (#1043).
     noVisionModels: OPENCODE_ZEN_TEXT_ONLY_MODELS,

package/src/providers/service-tier.ts

@@ -70,11 +70,22 @@ function buildFastPolicyAuthority(
   capabilityProvider: ServiceTierCapabilityProvider = provider,
 ): FastPolicyAuthority {
   const registry = registryTransportMatch ? getProviderRegistryEntry(providerName) : undefined;
+  const authTransport = resolveProviderAuthTransport(
+    provider.adapter,
+    provider.authMode ?? registry?.authKind ?? "key",
+    provider.apiKeyTransport,
+  );
+  const keyAuthDefaults = registry?.allowKeyAuthOverride === true
+    && (authTransport === "authorization_bearer" || authTransport === "x_api_key")
+    ? registry.keyAuthServiceTier
+    : undefined;
   const registryModelCapabilities = registry
     && registryModelServiceTierCapabilityApplies(registry, capabilityProvider)
     ? registry.modelSupportsServiceTier
     : undefined;
-  const providerCapability = capabilityProvider.supportsServiceTier ?? registry?.supportsServiceTier;
+  const providerCapability = capabilityProvider.supportsServiceTier
+    ?? keyAuthDefaults?.supportsServiceTier
+    ?? registry?.supportsServiceTier;
   const authority: FastPolicyAuthority = Object.freeze({
     providerAdapter: provider.adapter,
     providerAuthMode: provider.authMode ?? registry?.authKind ?? "key",
@@ -82,19 +93,23 @@ function buildFastPolicyAuthority(
       provider.fastWire !== undefined ? provider.fastWire : registry?.fastWire,
       { freeze: true },
     ),
+    ...(registry?.fastTierDescription !== undefined
+      ? { fastTierDescription: registry.fastTierDescription }
+      : {}),
     modelWireOverrideAllowed: !isCanonicalOpenAiForwardProvider(provider as OcxProviderConfig),
-    authTransport: resolveProviderAuthTransport(
-      provider.adapter,
-      provider.authMode ?? registry?.authKind ?? "key",
-      provider.apiKeyTransport,
-    ),
+    authTransport,
     capability: Object.freeze({
       ...(providerCapability !== undefined ? { provider: providerCapability } : {}),
       models: Object.freeze({
         ...(registryModelCapabilities ?? {}),
+        ...(keyAuthDefaults?.modelSupportsServiceTier ?? {}),
         ...(capabilityProvider.modelSupportsServiceTier ?? {}),
       }),
-      ...(provider.chatServiceTier !== undefined ? { chatServiceTier: provider.chatServiceTier } : {}),
+      ...(provider.chatServiceTier !== undefined
+        ? { chatServiceTier: provider.chatServiceTier }
+        : keyAuthDefaults?.chatServiceTier !== undefined
+          ? { chatServiceTier: keyAuthDefaults.chatServiceTier }
+          : {}),
     }),
     modelAdapters: Object.freeze({ ...(provider.modelAdapters ?? {}) }),
     hardPins: captureWireAdapterHardPins(providerName),

package/src/responses/custom-tool-compat.ts

@@ -4,6 +4,13 @@ import { collectResponsesToolGroups } from "./tool-groups";
 const ROUTED_CUSTOM_TOOL_PASSTHROUGH = new Set(["apply_patch"]);
 const BUILTIN_FUNCTIONS_NAMESPACE = "functions";
 
+function routedCustomToolPassesThrough(
+  name: string,
+  supportsResponsesCustomTools: boolean | undefined,
+): boolean {
+  return supportsResponsesCustomTools !== false && ROUTED_CUSTOM_TOOL_PASSTHROUGH.has(name);
+}
+
 function isPlainObject(value: unknown): value is Record<string, unknown> {
   return !!value && typeof value === "object" && !Array.isArray(value);
 }
@@ -34,7 +41,10 @@ export function routedCustomToolWireName(value: unknown): string | undefined {
  * Names of converted custom declarations after namespace lowering. Restoration uses these exact
  * wire identities so same-named function and custom children in different namespaces stay distinct.
  */
-function collectRoutedCustomToolWireNames(body: unknown): Set<string> {
+function collectRoutedCustomToolWireNames(
+  body: unknown,
+  supportsResponsesCustomTools?: boolean,
+): Set<string> {
   const names = new Set<string>();
   const groups = collectResponsesToolGroups(body);
   const bareWireNames = new Set<string>();
@@ -54,7 +64,7 @@ function collectRoutedCustomToolWireNames(body: unknown): Set<string> {
       if (
         tool.type === "custom"
         && typeof tool.name === "string"
-        && !ROUTED_CUSTOM_TOOL_PASSTHROUGH.has(tool.name)
+        && !routedCustomToolPassesThrough(tool.name, supportsResponsesCustomTools)
       ) {
         names.add(tool.name);
         continue;
@@ -67,7 +77,7 @@ function collectRoutedCustomToolWireNames(body: unknown): Set<string> {
           isPlainObject(child)
           && child.type === "custom"
           && typeof child.name === "string"
-          && !ROUTED_CUSTOM_TOOL_PASSTHROUGH.has(child.name)
+          && !routedCustomToolPassesThrough(child.name, supportsResponsesCustomTools)
           && !(tool.name === BUILTIN_FUNCTIONS_NAMESPACE && bareWireNames.has(child.name))
         ) names.add(customToolWireName(tool.name, child.name));
       }
@@ -81,7 +91,10 @@ export function customToolItemId(id: unknown): unknown {
   return id.startsWith("fc_") ? `ctc_${id.slice(3)}` : id;
 }
 
-export function collectRoutedCustomToolNames(body: unknown): Set<string> {
+export function collectRoutedCustomToolNames(
+  body: unknown,
+  supportsResponsesCustomTools?: boolean,
+): Set<string> {
   const names = new Set<string>();
   const visit = (value: unknown): void => {
     if (Array.isArray(value)) {
@@ -92,7 +105,7 @@ export function collectRoutedCustomToolNames(body: unknown): Set<string> {
     if (
       value.type === "custom"
       && typeof value.name === "string"
-      && !ROUTED_CUSTOM_TOOL_PASSTHROUGH.has(value.name)
+      && !routedCustomToolPassesThrough(value.name, supportsResponsesCustomTools)
     ) {
       names.add(value.name);
     }
@@ -184,12 +197,15 @@ function rewriteForUpstream(
   return changed ? next : value;
 }
 
-export function rewriteRoutedCustomToolsForUpstream(body: unknown): {
+export function rewriteRoutedCustomToolsForUpstream(
+  body: unknown,
+  supportsResponsesCustomTools?: boolean,
+): {
   body: unknown;
   names: Set<string>;
 } {
-  const conversionNames = collectRoutedCustomToolNames(body);
-  const names = collectRoutedCustomToolWireNames(body);
+  const conversionNames = collectRoutedCustomToolNames(body, supportsResponsesCustomTools);
+  const names = collectRoutedCustomToolWireNames(body, supportsResponsesCustomTools);
   if (conversionNames.size === 0) return { body, names };
   const callIds = new Set<string>();
   collectConvertedCallIds(body, conversionNames, callIds);