@gullabs/xai 0.4.1 → 0.5.0

package/dist/index.d.cts

@@ -99,7 +99,7 @@ interface XaiResponseCreateParams {
     input: XaiInputItem[];
     instructions?: string;
     reasoning?: {
-        effort: 'low' | 'high';
+        effort: 'low' | 'medium' | 'high' | 'xhigh';
     };
     text?: {
         format: XaiTextFormat;
@@ -108,6 +108,11 @@ interface XaiResponseCreateParams {
     top_p?: number;
     max_output_tokens?: number;
     prompt_cache_key?: string;
+    /**
+     * Responses `service_tier`. Only `'priority'` is forwarded (live-verified
+     * on grok-4.6, 2026-08-12). Omitted when the caller did not request a tier.
+     */
+    service_tier?: 'priority';
     /** Always `false` — this library never relies on xAI-side conversation storage. */
     store: false;
 }
@@ -181,6 +186,11 @@ interface XaiResponseShape {
         effort?: string;
         summary?: string;
     };
+    /**
+     * Echoed served tier. Observed values: `"default"`, `"priority"`.
+     * Kept as a plain string — xAI may add further values.
+     */
+    service_tier?: string;
     store?: boolean;
     prompt_cache_key?: string | null;
     /**
@@ -446,19 +456,53 @@ declare const Grok45ConfigSchema: z.ZodObject<{
     }, z.core.$strict>>;
 }, z.core.$strict>;
 
+/**
+ * Strict Zod config schema for xAI's `grok-4.6` model.
+ *
+ * Same Responses-API surface as grok-4.5, plus live-verified (2026-08-12)
+ * `reasoning.effort` of `'low' | 'medium' | 'high' | 'xhigh'` and
+ * `serviceTier: 'priority'`. `'none'` is rejected by the live API. Unknown
+ * tiers (`flex`, `standard`, `batch`) are rejected — `flex` is silently
+ * remapped to `default` by xAI, so this schema never admits it.
+ *
+ * @module
+ */
+
+declare const Grok46ConfigSchema: z.ZodObject<{
+    temperature: z.ZodOptional<z.ZodNumber>;
+    topP: z.ZodOptional<z.ZodNumber>;
+    maxOutputTokens: z.ZodOptional<z.ZodNumber>;
+    reasoning: z.ZodOptional<z.ZodObject<{
+        effort: z.ZodEnum<{
+            low: "low";
+            medium: "medium";
+            high: "high";
+            xhigh: "xhigh";
+        }>;
+    }, z.core.$strict>>;
+    serviceTier: z.ZodOptional<z.ZodLiteral<"priority">>;
+    timeoutMs: z.ZodOptional<z.ZodNumber>;
+    providerOptions: z.ZodOptional<z.ZodObject<{
+        xai: z.ZodOptional<z.ZodObject<{
+            promptCacheKey: z.ZodOptional<z.ZodString>;
+        }, z.core.$strict>>;
+    }, z.core.$strict>>;
+}, z.core.$strict>;
+
 /**
  * Model descriptor + registry for @gullabs/xai.
  *
- * v1 ships exactly one model: `grok-4.5` (canonical id only — the
- * `grok-4.5-latest` / `grok-build-latest` aliases visible in xAI's
- * `/v1/models` listing are intentionally NOT registered as separate
- * descriptors; reject-don't-map, callers must use `grok-4.5` verbatim).
+ * Ships two canonical models: `grok-4.5` and `grok-4.6`. xAI aliases
+ * (`grok-4.5-latest`, `grok-build-latest`) visible in `/v1/models` are
+ * intentionally NOT registered (reject-don't-map). `grok-4.6` has no
+ * aliases as of the 2026-08-12 `/v1/models` listing.
  *
  * @module
  */
 
 declare const grok45ModelDescriptor: ModelDescriptor;
-/** Every model descriptor `@gullabs/xai` contributes. v1 ships only grok-4.5. */
+declare const grok46ModelDescriptor: ModelDescriptor;
+/** Every model descriptor `@gullabs/xai` contributes. */
 declare const xaiModelDescriptors: ModelDescriptor[];
 declare const xaiRegistry: ModelRegistry;
 
@@ -474,35 +518,33 @@ declare const xaiRegistry: ModelRegistry;
  * (those are Gemini-only). `PricingSource` is provider-scoped by contract
  * (see `packages/core/src/ports.ts`); this module is xai's own.
  *
- * **Long-context tier.** grok-4.5 charges a premium when the GROSS input
- * token count exceeds 200,000 (`long_context_threshold` in xAI's
+ * **Long-context tier.** grok-4.5 / grok-4.6 charge a premium when the GROSS
+ * input token count exceeds 200,000 (`long_context_threshold` in xAI's
  * `/v1/models` listing). Selected by `inputTokens` (incl. cached), not by
  * billable input — mirrors core's `selectRates` convention exactly (strictly
  * greater than 200,000).
  *
- * **Service tiers.** xAI has no service-tier concept for grok-4.5 — there is
- * no `TIER_FACTOR`-equivalent here. `price()`'s `tier` param is accepted for
- * `PricingSource` structural conformance but any *defined* tier is treated
- * as unrecognized → unpriced (reject-don't-map), mirroring core's own
- * defensive stance for an unrecognized tier. `undefined` (no tier requested,
- * the only value xAI adapters ever pass — `xaiAdapter` rejects `serviceTier`
- * upstream) prices normally.
+ * **Service tiers.** grok-4.5 has none. grok-4.6 admits `'priority'`
+ * (echo live-verified 2026-08-12). The 2× multiplier is confirmed by
+ * fixture `12-grok-4-6-xhigh-priority.json` (`cost_in_usd_ticks` equals
+ * exactly 2× standard list). `'default'` (the value xAI echoes when no
+ * priority is served) and `undefined` (no tier requested) price at the
+ * standard list. Any other defined tier is unpriced (reject-don't-map).
  *
  * **Conversion factor.** xAI's `/v1/models` raw `*_token_price` fields are
  * in hundred-thousandths of a dollar per token (i.e. divide the raw integer
- * by 10,000 to get USD per million tokens): e.g. `grok-4.5`'s raw
- * `prompt_text_token_price: 20000` ÷ 10,000 = $2.00/M, which matches the
- * confirmed live-verified figure.
+ * by 10,000 to get USD per million tokens): e.g. `grok-4.6`'s raw
+ * `prompt_text_token_price: 20000` ÷ 10,000 = $2.00/M.
  *
- * Verified against `/v1/models` fixture captured 2026-07-09 (see
- * `docs/provider-plugins-and-xai-grok-4-5-plan.md` and the live-verification
- * fixture directory referenced in the commit-3 task brief).
+ * Verified against `/v1/models` on 2026-08-12. Prior snapshot
+ * `xai-2026-07-09` priced grok-4.5 cached input at $0.50 / $1.00; the live
+ * listing now reports $0.30 / $0.60.
  *
  * @module
  */
 
 /** Identifies this pricing snapshot — bump the date when rates change. */
-declare const xaiPricingVersion: "xai-2026-07-09";
+declare const xaiPricingVersion: "xai-2026-08-12";
 /**
  * Per-model rate entry (all values in µUSD per million tokens).
  *
@@ -521,6 +563,14 @@ interface XaiModelRates {
         cachedPerM: number;
         outputPerM: number;
     };
+    /**
+     * Multiplier for Responses `service_tier: "priority"`. Absent = this
+     * model does not admit priority (unpriced). Uncached standard-list 2×
+     * is confirmed by fixture `12-grok-4-6-xhigh-priority.json` ticks;
+     * cached and `gt200k` legs follow the official 2×-after-cache-discount
+     * docs rule (that fixture has cached=0 and input < 200k).
+     */
+    priorityFactor?: number;
 }
 /**
  * Frozen xAI pricing snapshot (per-1M in µUSD).
@@ -539,8 +589,10 @@ declare const XAI_PRICING: Readonly<Record<string, XaiModelRates>>;
  * **Algorithm** (mirrors `@gullabs/core`'s `computeCost` exactly, xai-owned):
  * 1. Look up rates for `model`; if not found, return an unpriced `Cost`
  *    (`microUsd: null`) naming the model.
- * 2. A defined `tier` is always unpriced (xai has no tiers; reject-don't-map).
- *    `undefined` prices normally.
+ * 2. `undefined` or `'default'` prices at the standard list.
+ *    `'priority'` applies `rates.priorityFactor` when present.
+ *    Any other defined tier, or `'priority'` on a model without
+ *    `priorityFactor`, is unpriced (reject-don't-map).
  * 3. Select base vs. `>200k` long-context rates from GROSS `inputTokens`.
  * 4. Billable input = `inputTokens − (cachedInputTokens ?? 0)`, clamped to 0.
  * 5. Round each component (input, cached, output) independently to the
@@ -558,7 +610,7 @@ declare function computeXaiCost(model: string, usage: Usage, tier?: string): Cos
  * import { xaiPricingSource } from '@gullabs/xai'
  *
  * const pricing = xaiPricingSource()
- * const cost = pricing.price('grok-4.5', usage)
+ * const cost = pricing.price('grok-4.6', usage)
  * ```
  */
 declare function xaiPricingSource(): PricingSource;
@@ -566,8 +618,9 @@ declare function xaiPricingSource(): PricingSource;
 /**
  * `xaiProvider` — {@link ProviderPlugin} factory for @gullabs/xai.
  *
- * Bundles the xAI Grok adapter, the `grok-4.5` model descriptor, and the
- * xai pricing source into a single plugin for {@link composeProviders}.
+ * Bundles the xAI Grok adapter, the `grok-4.5` / `grok-4.6` model
+ * descriptors, and the xai pricing source into a single plugin for
+ * {@link composeProviders}.
  *
  * @module
  */
@@ -576,8 +629,8 @@ declare function xaiPricingSource(): PricingSource;
  * Create a {@link ProviderPlugin} for the xAI Grok provider.
  *
  * @param opts - Forwarded to {@link xaiAdapter}.
- * @returns A plugin bundling the xAI adapter, the `grok-4.5` model
- *   descriptor, and the built-in xai pricing source.
+ * @returns A plugin bundling the xAI adapter, the `grok-4.5` / `grok-4.6`
+ *   model descriptors, and the built-in xai pricing source.
  *
  * @example
  * ```ts
@@ -591,4 +644,4 @@ declare function xaiPricingSource(): PricingSource;
  */
 declare function xaiProvider(opts?: XaiAdapterOptions): ProviderPlugin;
 
-export { type FileDeleteOptions, Grok45ConfigSchema, XAI_FILES_DEFAULT_BASE_URL, XAI_FILE_MAX_BYTES, XAI_FILE_TTL_MAX_SECONDS, XAI_FILE_TTL_MIN_SECONDS, XAI_PRICING, type XaiAdapterOptions, type XaiClientLike, type XaiFileHandle, type XaiFileListOptions, type XaiFileListResult, XaiFileStore, type XaiFileStoreOptions, type XaiFileUploadInput, type XaiInputContentPart, type XaiInputFilePart, type XaiInputImagePart, type XaiInputItem, type XaiInputTextPart, type XaiMessageOutputItem, type XaiModelRates, type XaiOutputItem, type XaiOutputTextPart, type XaiProviderOptions, type XaiReasoningOutputItem, type XaiReasoningSummaryPart, type XaiResponseCreateParams, type XaiResponseShape, type XaiTextFormat, type XaiUsageShape, buildXaiClient, classifyXaiError, computeXaiCost, grok45ModelDescriptor, requireApiKey, xaiAdapter, xaiModelDescriptors, xaiPricingSource, xaiPricingVersion, xaiProvider, xaiRegistry };
+export { type FileDeleteOptions, Grok45ConfigSchema, Grok46ConfigSchema, XAI_FILES_DEFAULT_BASE_URL, XAI_FILE_MAX_BYTES, XAI_FILE_TTL_MAX_SECONDS, XAI_FILE_TTL_MIN_SECONDS, XAI_PRICING, type XaiAdapterOptions, type XaiClientLike, type XaiFileHandle, type XaiFileListOptions, type XaiFileListResult, XaiFileStore, type XaiFileStoreOptions, type XaiFileUploadInput, type XaiInputContentPart, type XaiInputFilePart, type XaiInputImagePart, type XaiInputItem, type XaiInputTextPart, type XaiMessageOutputItem, type XaiModelRates, type XaiOutputItem, type XaiOutputTextPart, type XaiProviderOptions, type XaiReasoningOutputItem, type XaiReasoningSummaryPart, type XaiResponseCreateParams, type XaiResponseShape, type XaiTextFormat, type XaiUsageShape, buildXaiClient, classifyXaiError, computeXaiCost, grok45ModelDescriptor, grok46ModelDescriptor, requireApiKey, xaiAdapter, xaiModelDescriptors, xaiPricingSource, xaiPricingVersion, xaiProvider, xaiRegistry };

package/dist/index.d.ts

@@ -99,7 +99,7 @@ interface XaiResponseCreateParams {
     input: XaiInputItem[];
     instructions?: string;
     reasoning?: {
-        effort: 'low' | 'high';
+        effort: 'low' | 'medium' | 'high' | 'xhigh';
     };
     text?: {
         format: XaiTextFormat;
@@ -108,6 +108,11 @@ interface XaiResponseCreateParams {
     top_p?: number;
     max_output_tokens?: number;
     prompt_cache_key?: string;
+    /**
+     * Responses `service_tier`. Only `'priority'` is forwarded (live-verified
+     * on grok-4.6, 2026-08-12). Omitted when the caller did not request a tier.
+     */
+    service_tier?: 'priority';
     /** Always `false` — this library never relies on xAI-side conversation storage. */
     store: false;
 }
@@ -181,6 +186,11 @@ interface XaiResponseShape {
         effort?: string;
         summary?: string;
     };
+    /**
+     * Echoed served tier. Observed values: `"default"`, `"priority"`.
+     * Kept as a plain string — xAI may add further values.
+     */
+    service_tier?: string;
     store?: boolean;
     prompt_cache_key?: string | null;
     /**
@@ -446,19 +456,53 @@ declare const Grok45ConfigSchema: z.ZodObject<{
     }, z.core.$strict>>;
 }, z.core.$strict>;
 
+/**
+ * Strict Zod config schema for xAI's `grok-4.6` model.
+ *
+ * Same Responses-API surface as grok-4.5, plus live-verified (2026-08-12)
+ * `reasoning.effort` of `'low' | 'medium' | 'high' | 'xhigh'` and
+ * `serviceTier: 'priority'`. `'none'` is rejected by the live API. Unknown
+ * tiers (`flex`, `standard`, `batch`) are rejected — `flex` is silently
+ * remapped to `default` by xAI, so this schema never admits it.
+ *
+ * @module
+ */
+
+declare const Grok46ConfigSchema: z.ZodObject<{
+    temperature: z.ZodOptional<z.ZodNumber>;
+    topP: z.ZodOptional<z.ZodNumber>;
+    maxOutputTokens: z.ZodOptional<z.ZodNumber>;
+    reasoning: z.ZodOptional<z.ZodObject<{
+        effort: z.ZodEnum<{
+            low: "low";
+            medium: "medium";
+            high: "high";
+            xhigh: "xhigh";
+        }>;
+    }, z.core.$strict>>;
+    serviceTier: z.ZodOptional<z.ZodLiteral<"priority">>;
+    timeoutMs: z.ZodOptional<z.ZodNumber>;
+    providerOptions: z.ZodOptional<z.ZodObject<{
+        xai: z.ZodOptional<z.ZodObject<{
+            promptCacheKey: z.ZodOptional<z.ZodString>;
+        }, z.core.$strict>>;
+    }, z.core.$strict>>;
+}, z.core.$strict>;
+
 /**
  * Model descriptor + registry for @gullabs/xai.
  *
- * v1 ships exactly one model: `grok-4.5` (canonical id only — the
- * `grok-4.5-latest` / `grok-build-latest` aliases visible in xAI's
- * `/v1/models` listing are intentionally NOT registered as separate
- * descriptors; reject-don't-map, callers must use `grok-4.5` verbatim).
+ * Ships two canonical models: `grok-4.5` and `grok-4.6`. xAI aliases
+ * (`grok-4.5-latest`, `grok-build-latest`) visible in `/v1/models` are
+ * intentionally NOT registered (reject-don't-map). `grok-4.6` has no
+ * aliases as of the 2026-08-12 `/v1/models` listing.
  *
  * @module
  */
 
 declare const grok45ModelDescriptor: ModelDescriptor;
-/** Every model descriptor `@gullabs/xai` contributes. v1 ships only grok-4.5. */
+declare const grok46ModelDescriptor: ModelDescriptor;
+/** Every model descriptor `@gullabs/xai` contributes. */
 declare const xaiModelDescriptors: ModelDescriptor[];
 declare const xaiRegistry: ModelRegistry;
 
@@ -474,35 +518,33 @@ declare const xaiRegistry: ModelRegistry;
  * (those are Gemini-only). `PricingSource` is provider-scoped by contract
  * (see `packages/core/src/ports.ts`); this module is xai's own.
  *
- * **Long-context tier.** grok-4.5 charges a premium when the GROSS input
- * token count exceeds 200,000 (`long_context_threshold` in xAI's
+ * **Long-context tier.** grok-4.5 / grok-4.6 charge a premium when the GROSS
+ * input token count exceeds 200,000 (`long_context_threshold` in xAI's
  * `/v1/models` listing). Selected by `inputTokens` (incl. cached), not by
  * billable input — mirrors core's `selectRates` convention exactly (strictly
  * greater than 200,000).
  *
- * **Service tiers.** xAI has no service-tier concept for grok-4.5 — there is
- * no `TIER_FACTOR`-equivalent here. `price()`'s `tier` param is accepted for
- * `PricingSource` structural conformance but any *defined* tier is treated
- * as unrecognized → unpriced (reject-don't-map), mirroring core's own
- * defensive stance for an unrecognized tier. `undefined` (no tier requested,
- * the only value xAI adapters ever pass — `xaiAdapter` rejects `serviceTier`
- * upstream) prices normally.
+ * **Service tiers.** grok-4.5 has none. grok-4.6 admits `'priority'`
+ * (echo live-verified 2026-08-12). The 2× multiplier is confirmed by
+ * fixture `12-grok-4-6-xhigh-priority.json` (`cost_in_usd_ticks` equals
+ * exactly 2× standard list). `'default'` (the value xAI echoes when no
+ * priority is served) and `undefined` (no tier requested) price at the
+ * standard list. Any other defined tier is unpriced (reject-don't-map).
  *
  * **Conversion factor.** xAI's `/v1/models` raw `*_token_price` fields are
  * in hundred-thousandths of a dollar per token (i.e. divide the raw integer
- * by 10,000 to get USD per million tokens): e.g. `grok-4.5`'s raw
- * `prompt_text_token_price: 20000` ÷ 10,000 = $2.00/M, which matches the
- * confirmed live-verified figure.
+ * by 10,000 to get USD per million tokens): e.g. `grok-4.6`'s raw
+ * `prompt_text_token_price: 20000` ÷ 10,000 = $2.00/M.
  *
- * Verified against `/v1/models` fixture captured 2026-07-09 (see
- * `docs/provider-plugins-and-xai-grok-4-5-plan.md` and the live-verification
- * fixture directory referenced in the commit-3 task brief).
+ * Verified against `/v1/models` on 2026-08-12. Prior snapshot
+ * `xai-2026-07-09` priced grok-4.5 cached input at $0.50 / $1.00; the live
+ * listing now reports $0.30 / $0.60.
  *
  * @module
  */
 
 /** Identifies this pricing snapshot — bump the date when rates change. */
-declare const xaiPricingVersion: "xai-2026-07-09";
+declare const xaiPricingVersion: "xai-2026-08-12";
 /**
  * Per-model rate entry (all values in µUSD per million tokens).
  *
@@ -521,6 +563,14 @@ interface XaiModelRates {
         cachedPerM: number;
         outputPerM: number;
     };
+    /**
+     * Multiplier for Responses `service_tier: "priority"`. Absent = this
+     * model does not admit priority (unpriced). Uncached standard-list 2×
+     * is confirmed by fixture `12-grok-4-6-xhigh-priority.json` ticks;
+     * cached and `gt200k` legs follow the official 2×-after-cache-discount
+     * docs rule (that fixture has cached=0 and input < 200k).
+     */
+    priorityFactor?: number;
 }
 /**
  * Frozen xAI pricing snapshot (per-1M in µUSD).
@@ -539,8 +589,10 @@ declare const XAI_PRICING: Readonly<Record<string, XaiModelRates>>;
  * **Algorithm** (mirrors `@gullabs/core`'s `computeCost` exactly, xai-owned):
  * 1. Look up rates for `model`; if not found, return an unpriced `Cost`
  *    (`microUsd: null`) naming the model.
- * 2. A defined `tier` is always unpriced (xai has no tiers; reject-don't-map).
- *    `undefined` prices normally.
+ * 2. `undefined` or `'default'` prices at the standard list.
+ *    `'priority'` applies `rates.priorityFactor` when present.
+ *    Any other defined tier, or `'priority'` on a model without
+ *    `priorityFactor`, is unpriced (reject-don't-map).
  * 3. Select base vs. `>200k` long-context rates from GROSS `inputTokens`.
  * 4. Billable input = `inputTokens − (cachedInputTokens ?? 0)`, clamped to 0.
  * 5. Round each component (input, cached, output) independently to the
@@ -558,7 +610,7 @@ declare function computeXaiCost(model: string, usage: Usage, tier?: string): Cos
  * import { xaiPricingSource } from '@gullabs/xai'
  *
  * const pricing = xaiPricingSource()
- * const cost = pricing.price('grok-4.5', usage)
+ * const cost = pricing.price('grok-4.6', usage)
  * ```
  */
 declare function xaiPricingSource(): PricingSource;
@@ -566,8 +618,9 @@ declare function xaiPricingSource(): PricingSource;
 /**
  * `xaiProvider` — {@link ProviderPlugin} factory for @gullabs/xai.
  *
- * Bundles the xAI Grok adapter, the `grok-4.5` model descriptor, and the
- * xai pricing source into a single plugin for {@link composeProviders}.
+ * Bundles the xAI Grok adapter, the `grok-4.5` / `grok-4.6` model
+ * descriptors, and the xai pricing source into a single plugin for
+ * {@link composeProviders}.
  *
  * @module
  */
@@ -576,8 +629,8 @@ declare function xaiPricingSource(): PricingSource;
  * Create a {@link ProviderPlugin} for the xAI Grok provider.
  *
  * @param opts - Forwarded to {@link xaiAdapter}.
- * @returns A plugin bundling the xAI adapter, the `grok-4.5` model
- *   descriptor, and the built-in xai pricing source.
+ * @returns A plugin bundling the xAI adapter, the `grok-4.5` / `grok-4.6`
+ *   model descriptors, and the built-in xai pricing source.
  *
  * @example
  * ```ts
@@ -591,4 +644,4 @@ declare function xaiPricingSource(): PricingSource;
  */
 declare function xaiProvider(opts?: XaiAdapterOptions): ProviderPlugin;
 
-export { type FileDeleteOptions, Grok45ConfigSchema, XAI_FILES_DEFAULT_BASE_URL, XAI_FILE_MAX_BYTES, XAI_FILE_TTL_MAX_SECONDS, XAI_FILE_TTL_MIN_SECONDS, XAI_PRICING, type XaiAdapterOptions, type XaiClientLike, type XaiFileHandle, type XaiFileListOptions, type XaiFileListResult, XaiFileStore, type XaiFileStoreOptions, type XaiFileUploadInput, type XaiInputContentPart, type XaiInputFilePart, type XaiInputImagePart, type XaiInputItem, type XaiInputTextPart, type XaiMessageOutputItem, type XaiModelRates, type XaiOutputItem, type XaiOutputTextPart, type XaiProviderOptions, type XaiReasoningOutputItem, type XaiReasoningSummaryPart, type XaiResponseCreateParams, type XaiResponseShape, type XaiTextFormat, type XaiUsageShape, buildXaiClient, classifyXaiError, computeXaiCost, grok45ModelDescriptor, requireApiKey, xaiAdapter, xaiModelDescriptors, xaiPricingSource, xaiPricingVersion, xaiProvider, xaiRegistry };
+export { type FileDeleteOptions, Grok45ConfigSchema, Grok46ConfigSchema, XAI_FILES_DEFAULT_BASE_URL, XAI_FILE_MAX_BYTES, XAI_FILE_TTL_MAX_SECONDS, XAI_FILE_TTL_MIN_SECONDS, XAI_PRICING, type XaiAdapterOptions, type XaiClientLike, type XaiFileHandle, type XaiFileListOptions, type XaiFileListResult, XaiFileStore, type XaiFileStoreOptions, type XaiFileUploadInput, type XaiInputContentPart, type XaiInputFilePart, type XaiInputImagePart, type XaiInputItem, type XaiInputTextPart, type XaiMessageOutputItem, type XaiModelRates, type XaiOutputItem, type XaiOutputTextPart, type XaiProviderOptions, type XaiReasoningOutputItem, type XaiReasoningSummaryPart, type XaiResponseCreateParams, type XaiResponseShape, type XaiTextFormat, type XaiUsageShape, buildXaiClient, classifyXaiError, computeXaiCost, grok45ModelDescriptor, grok46ModelDescriptor, requireApiKey, xaiAdapter, xaiModelDescriptors, xaiPricingSource, xaiPricingVersion, xaiProvider, xaiRegistry };

package/dist/index.js

@@ -260,10 +260,19 @@ function xaiAdapter(opts) {
       if (genConfig.maxOutputTokens !== void 0) {
         params.max_output_tokens = genConfig.maxOutputTokens;
       }
+      const admittedTiers = req.modelDescriptor?.capabilities?.serviceTiers;
       if (genConfig.serviceTier !== void 0) {
-        throw badXaiRequest(
-          `serviceTier is not supported for xai models (got "${genConfig.serviceTier}").`
-        );
+        if (admittedTiers === void 0 || !admittedTiers.includes(genConfig.serviceTier)) {
+          throw badXaiRequest(
+            `serviceTier is not supported for xai model "${model}" (got "${genConfig.serviceTier}").`
+          );
+        }
+        if (genConfig.serviceTier !== "priority") {
+          throw badXaiRequest(
+            `serviceTier "${genConfig.serviceTier}" is not supported for xai model "${model}" (only "priority" is admitted).`
+          );
+        }
+        params.service_tier = "priority";
       }
       const reasoning = genConfig.reasoning;
       if (reasoning !== void 0) {
@@ -274,13 +283,13 @@ function xaiAdapter(opts) {
         }
         if (reasoning.effort !== void 0) {
           const effort = reasoning.effort;
-          if (effort !== "low" && effort !== "high") {
+          if (effort === "none") {
             throw badXaiRequest(
-              `reasoning.effort "${effort}" is not supported for xai model "${model}" (only "low" and "high" are admitted).`
+              `reasoning.effort "none" is not supported for xai model "${model}".`
             );
           }
           const admitted = req.modelDescriptor?.capabilities?.admittedReasoningEfforts;
-          if (admitted !== void 0 && !admitted.includes(effort)) {
+          if (admitted === void 0 || !admitted.includes(effort)) {
             throw badXaiRequest(
               `reasoning.effort "${effort}" is not supported for xai model "${model}".`
             );
@@ -358,6 +367,7 @@ function xaiAdapter(opts) {
       if (isPlainRecord(response.metadata)) {
         providerMeta["metadata"] = response.metadata;
       }
+      const servedServiceTier = typeof response.service_tier === "string" && response.service_tier.length > 0 ? response.service_tier : void 0;
       const result = {
         model: response.model,
         usage,
@@ -367,6 +377,7 @@ function xaiAdapter(opts) {
         ...text.length > 0 ? { text } : {},
         ...reasoningText !== void 0 ? { reasoningText } : {},
         ...rawStructured !== void 0 ? { rawStructured } : {},
+        ...servedServiceTier !== void 0 ? { servedServiceTier } : {},
         ...Object.keys(providerMeta).length > 0 ? { providerMetadata: providerMeta } : {}
       };
       return result;
@@ -873,6 +884,55 @@ var Grok45ConfigSchema = z.strictObject({
   description: "Strict Responses API config for model grok-4.5. Level reasoning (low/high only), tunable sampling, no service tiers, structured output, vision, priced.",
   examples: [{ reasoning: { effort: "high" } }]
 });
+var Grok46ConfigSchema = z.strictObject({
+  temperature: z.number().optional().meta({
+    title: "Temperature",
+    description: "Sampling temperature forwarded verbatim to grok-4.6."
+  }),
+  topP: z.number().optional().meta({
+    title: "Top P",
+    description: "Nucleus sampling parameter forwarded verbatim to grok-4.6."
+  }),
+  maxOutputTokens: z.number().int().positive().optional().meta({
+    title: "Max Output Tokens",
+    description: "Maximum output token cap for grok-4.6. No artificial ceiling \u2014 xAI accepts arbitrarily large values; truncation surfaces as finishReason:'length', not an error."
+  }),
+  reasoning: z.strictObject({
+    effort: z.enum(["low", "medium", "high", "xhigh"]).meta({
+      title: "Reasoning Effort",
+      description: 'Reasoning effort for grok-4.6. Live-verified 2026-08-12: "low", "medium", "high", and "xhigh" are accepted; "none" is rejected. Vendor default when omitted is "high".'
+    })
+  }).optional().meta({
+    title: "Reasoning",
+    description: "grok-4.6 effort-level reasoning configuration. No budgetTokens field \u2014 xAI uses level-style reasoning, not token budgets."
+  }),
+  serviceTier: z.literal("priority").optional().meta({
+    title: "Service Tier",
+    description: 'xAI priority processing for grok-4.6 (Responses `service_tier: "priority"`). Echo live-verified 2026-08-12. Bills at 2\xD7 after the cache discount (uncached standard-list 2\xD7 confirmed by live ticks; cached/long-context legs follow the official 2\xD7 rule). Omitted requests stay on xAI default. "flex"/"standard"/"batch" are rejected.'
+  }),
+  timeoutMs: z.number().int().positive().optional().meta({
+    title: "Timeout",
+    description: "Logical request timeout in milliseconds."
+  }),
+  providerOptions: z.strictObject({
+    xai: z.strictObject({
+      promptCacheKey: z.string().min(1).optional().meta({
+        title: "Prompt Cache Key",
+        description: "xAI conversation-routing cache key \u2014 maps to Responses API `prompt_cache_key`."
+      })
+    }).optional().meta({
+      title: "xAI Provider Options",
+      description: "Allowlisted xAI provider options for grok-4.6."
+    })
+  }).optional().meta({
+    title: "Provider Options",
+    description: "Provider-specific options accepted for grok-4.6."
+  })
+}).meta({
+  title: "Grok46Config",
+  description: "Strict Responses API config for model grok-4.6. Level reasoning (low/medium/high/xhigh), optional priority service tier, tunable sampling, structured output, vision, priced.",
+  examples: [{ reasoning: { effort: "high" } }]
+});
 
 // src/models.ts
 var grok45ModelDescriptor = {
@@ -890,20 +950,55 @@ var grok45ModelDescriptor = {
     sampling: "tunable",
     caching: { explicit: false, minTokens: 0 },
     grounding: false
-    // No serviceTiers key — xai has no service-tier concept.
+    // No serviceTiers key — grok-4.5 has no admitted service-tier vocabulary.
   },
   configSchema: Grok45ConfigSchema,
   configJsonSchema: toConfigJsonSchema(Grok45ConfigSchema),
   validateConfig: zodToStandardSchema(Grok45ConfigSchema)
 };
-var xaiModelDescriptors = [grok45ModelDescriptor];
+var grok46ModelDescriptor = {
+  model: "grok-4.6",
+  provider: "xai",
+  pricingFamily: "grok-4.6",
+  capabilities: {
+    reasoning: true,
+    reasoningApi: "level",
+    admittedReasoningEfforts: ["low", "medium", "high", "xhigh"],
+    structuredOutput: true,
+    nativeStructuredOutput: true,
+    vision: true,
+    audioInput: false,
+    sampling: "tunable",
+    caching: { explicit: false, minTokens: 0 },
+    grounding: false,
+    serviceTiers: ["priority"]
+  },
+  configSchema: Grok46ConfigSchema,
+  configJsonSchema: toConfigJsonSchema(Grok46ConfigSchema),
+  validateConfig: zodToStandardSchema(Grok46ConfigSchema)
+};
+var xaiModelDescriptors = [
+  grok45ModelDescriptor,
+  grok46ModelDescriptor
+];
 var xaiRegistry = createModelRegistry(xaiModelDescriptors);
 
 // src/pricing.ts
-var xaiPricingVersion = "xai-2026-07-09";
+var xaiPricingVersion = "xai-2026-08-12";
 var XAI_PRICING = Object.freeze({
-  // ── grok-4.5 ──  $2.00/$6.00 (≤200k), $4.00/$12.00 (>200k); cached $0.50/$1.00
+  // ── grok-4.5 ──  $2.00/$6.00 (≤200k), $4.00/$12.00 (>200k); cached $0.30/$0.60
   "grok-4.5": {
+    inputPerM: 2e6,
+    cachedPerM: 3e5,
+    outputPerM: 6e6,
+    gt200k: {
+      inputPerM: 4e6,
+      cachedPerM: 6e5,
+      outputPerM: 12e6
+    }
+  },
+  // ── grok-4.6 ──  $2.00/$6.00 (≤200k), $4.00/$12.00 (>200k); cached $0.50/$1.00
+  "grok-4.6": {
     inputPerM: 2e6,
     cachedPerM: 5e5,
     outputPerM: 6e6,
@@ -911,7 +1006,9 @@ var XAI_PRICING = Object.freeze({
       inputPerM: 4e6,
       cachedPerM: 1e6,
       outputPerM: 12e6
-    }
+    },
+    // Confirmed 2026-08-12 by fixture 12 cost_in_usd_ticks (2× list).
+    priorityFactor: 2
   }
 });
 var LONG_CONTEXT_THRESHOLD = 2e5;
@@ -940,22 +1037,29 @@ function computeXaiCost(model, usage, tier) {
       unpricedReason: `Unknown model "${model}"; no pricing entry found.`
     };
   }
-  if (tier !== void 0) {
-    return {
-      microUsd: null,
-      usd: null,
-      pricingVersion: xaiPricingVersion,
-      confidence: "estimated",
-      details: { input: 0, cached: 0, output: 0 },
-      unpricedReason: `Unknown service tier "${tier}"; xai has no service tiers, refusing to guess a pricing multiplier.`
-    };
+  let factor = 1;
+  if (tier !== void 0 && tier !== "default") {
+    if (tier === "priority" && rates.priorityFactor !== void 0) {
+      factor = rates.priorityFactor;
+    } else {
+      return {
+        microUsd: null,
+        usd: null,
+        pricingVersion: xaiPricingVersion,
+        confidence: "estimated",
+        details: { input: 0, cached: 0, output: 0 },
+        unpricedReason: `Unknown service tier "${tier}"; xai model "${model}" has no such tier, refusing to guess a pricing multiplier.`
+      };
+    }
   }
   const base = selectRates(rates, usage.inputTokens);
   const cached = usage.cachedInputTokens ?? 0;
   const billableInput = Math.max(0, usage.inputTokens - cached);
-  const inputCost = Math.round(billableInput * base.inputPerM / 1e6);
-  const cachedCost = Math.round(cached * base.cachedPerM / 1e6);
-  const outputCost = Math.round(usage.outputTokens * base.outputPerM / 1e6);
+  const inputCost = Math.round(billableInput * base.inputPerM * factor / 1e6);
+  const cachedCost = Math.round(cached * base.cachedPerM * factor / 1e6);
+  const outputCost = Math.round(
+    usage.outputTokens * base.outputPerM * factor / 1e6
+  );
   const microUsd = inputCost + cachedCost + outputCost;
   return {
     microUsd,
@@ -993,6 +1097,6 @@ function xaiProvider(opts) {
   };
 }
 
-export { Grok45ConfigSchema, XAI_FILES_DEFAULT_BASE_URL, XAI_FILE_MAX_BYTES, XAI_FILE_TTL_MAX_SECONDS, XAI_FILE_TTL_MIN_SECONDS, XAI_PRICING, XaiFileStore, buildXaiClient, classifyXaiError, computeXaiCost, grok45ModelDescriptor, requireApiKey, xaiAdapter, xaiModelDescriptors, xaiPricingSource, xaiPricingVersion, xaiProvider, xaiRegistry };
+export { Grok45ConfigSchema, Grok46ConfigSchema, XAI_FILES_DEFAULT_BASE_URL, XAI_FILE_MAX_BYTES, XAI_FILE_TTL_MAX_SECONDS, XAI_FILE_TTL_MIN_SECONDS, XAI_PRICING, XaiFileStore, buildXaiClient, classifyXaiError, computeXaiCost, grok45ModelDescriptor, grok46ModelDescriptor, requireApiKey, xaiAdapter, xaiModelDescriptors, xaiPricingSource, xaiPricingVersion, xaiProvider, xaiRegistry };
 //# sourceMappingURL=index.js.map
 //# sourceMappingURL=index.js.map