@jaypie/llm 1.3.15 → 1.3.17

package/dist/cjs/index.d.cts

@@ -21,6 +21,8 @@ declare const EFFORT: {
 };
 type LlmEffort = (typeof EFFORT)[keyof typeof EFFORT];
 declare const MODEL: {
+    NOVA_LITE: string;
+    NOVA_PRO: string;
     FABLE: string;
     OPUS: string;
     SONNET: string;
@@ -29,6 +31,7 @@ declare const MODEL: {
     FIREWORKS: {
         DEEPSEEK: string;
         GLM: string;
+        GPT_OSS: string;
         KIMI: string;
         MINIMAX: string;
         QWEN: string;
@@ -52,17 +55,78 @@ declare const MODEL: {
         SONNET: string;
     };
 };
+/** Price of one million tokens, in US dollars. */
+interface LlmModelCost {
+    /**
+     * Cache-read (cached input) tokens. Omitted when the provider does not price
+     * cache reads separately from uncached input.
+     */
+    cachedInputRead?: number;
+    /**
+     * Cache-write tokens. A scalar when the rate does not vary by TTL (`0` where
+     * the provider publishes a free write); keyed by the same `"5m"` / `"1h"`
+     * literals as `LlmCache` when it does, so a cost calculation reads the TTL
+     * straight off `OperateRequest.cache`. Omitted means cache writes bill at the
+     * `input` rate.
+     */
+    cachedInputWrite?: number | {
+        "1h": number;
+        "5m": number;
+    };
+    /** Uncached input tokens. */
+    input: number;
+    /** Output tokens. */
+    output: number;
+    /**
+     * Reasoning tokens, when billed at a rate other than `output`. Omitted means
+     * reasoning bills as output, which is true of every provider listed here.
+     */
+    reasoning?: number;
+}
+/**
+ * Standard list price per million tokens, keyed by literal model id (verified
+ * 2026-07-21). Keys are string literals rather than `MODEL.*` references so a
+ * model retired from the catalog keeps its price here: historic ids stay
+ * priceable after they leave `MODEL.*`, which is what makes replaying old
+ * usage records possible.
+ *
+ * Caveats the numbers cannot carry:
+ * - **Standard rate only.** Introductory, batch, flex, priority, fast-mode, and
+ *   data-residency rates are excluded. Sonnet 5 is listed at its standard
+ *   $3/$15, not the introductory rate.
+ * - **Cache writes are Anthropic-only.** Bedrock publishes a literal $0 write
+ *   for Amazon's own models, recorded here as `cachedInputWrite: 0`. Fireworks
+ *   writes bill at the input rate. OpenAI and xAI discount reads automatically
+ *   and publish no write premium. Google charges nothing to write an implicit
+ *   cache; explicit caching bills storage per hour, a unit this table does not
+ *   carry (and one `@jaypie/llm` does not wire).
+ * - **Short-context tier.** Long-prompt surcharges are not modeled: Gemini 3.1
+ *   Pro doubles above 200K, Grok doubles at 200K, GPT-5.5 is 2x in / 1.5x out
+ *   above 272K.
+ * - **Text rates.** Gemini prices audio input higher than text/image/video.
+ * - **Proxy routes are deliberately absent.** `MODEL.OPENROUTER.*`, and any
+ *   uncatalogued Bedrock id reselling a third-party model, are another vendor's
+ *   model behind a gateway, priced per route and per region, so no single rate
+ *   here would be correct. Price those against the backend model, or against
+ *   the gateway's own published rate. Amazon's own Nova models are first-class,
+ *   not proxies, and are priced below at the standard US on-demand rate (`us.`
+ *   geo profile for Nova 2 Lite; the `global.` profile is cheaper and is not
+ *   modeled).
+ *
+ * Unlisted ids return `undefined` — callers must handle a miss.
+ */
+declare const COST: Record<string, LlmModelCost>;
 declare const PROVIDER: {
     readonly BEDROCK: {
-        readonly DEFAULT: "amazon.nova-pro-v1:0";
-        /** @deprecated Size tiers are retired in 2.0. Use PROVIDER.BEDROCK.DEFAULT. */
+        readonly DEFAULT: string;
+        /** @deprecated Size tiers are retired in 2.0. Use PROVIDER.BEDROCK.DEFAULT, or pick a specific model from MODEL.NOVA_*. */
         readonly MODEL: {
-            readonly DEFAULT: "amazon.nova-lite-v1:0";
-            readonly LARGE: "amazon.nova-pro-v1:0";
-            readonly SMALL: "amazon.nova-lite-v1:0";
-            readonly TINY: "amazon.nova-micro-v1:0";
+            readonly DEFAULT: string;
+            readonly LARGE: string;
+            readonly SMALL: string;
+            readonly TINY: string;
         };
-        readonly MODEL_MATCH_WORDS: readonly ["amazon.nova", "amazon.titan", "anthropic.claude", "cohere.command", "meta.llama", "mistral.mistral", "ai21."];
+        readonly MODEL_MATCH_WORDS: readonly ["ai21.", "amazon.nova", "amazon.titan", "anthropic.claude", "cohere.command", "deepseek.", "google.gemma", "meta.llama", "mistral.mistral", "moonshotai.", "openai.gpt-oss", "qwen."];
         readonly NAME: "bedrock";
         readonly REGION: "AWS_REGION";
     };
@@ -206,22 +270,24 @@ declare const DEFAULT: {
  */
 declare const ALL: {
     readonly BASE: readonly ["claude-sonnet-4-6", "gemini-3.1-pro-preview", "gpt-5.4", "grok-latest"];
-    readonly COMBINED: readonly ("claude-opus-4-8" | "claude-haiku-4-5" | "gemini-3.5-flash" | "gemini-3.1-flash-lite" | "gemini-3.1-pro-preview" | "gpt-5.5" | "gpt-5.4-mini" | "gpt-5.4-nano" | "grok-latest" | "claude-sonnet-4-6" | "gpt-5.4" | "grok-4-1-fast-reasoning" | "grok-4-1-fast-non-reasoning")[];
+    readonly COMBINED: readonly ("claude-opus-4-8" | "claude-haiku-4-5" | "gemini-3.1-pro-preview" | "gpt-5.5" | "gpt-5.4-mini" | "gpt-5.4-nano" | "grok-latest" | "claude-sonnet-4-6" | "gemini-3.5-flash" | "gemini-3.1-flash-lite" | "gpt-5.4" | "grok-4-1-fast-reasoning" | "grok-4-1-fast-non-reasoning")[];
     readonly LARGE: readonly ["claude-opus-4-8", "gemini-3.1-pro-preview", "gpt-5.5", "grok-latest"];
     readonly SMALL: readonly ["claude-sonnet-4-6", "gemini-3.5-flash", "gpt-5.4-mini", "grok-4-1-fast-reasoning"];
     readonly TINY: readonly ["claude-haiku-4-5", "gemini-3.1-flash-lite", "gpt-5.4-nano", "grok-4-1-fast-non-reasoning"];
 };
 
 declare const constants_ALL: typeof ALL;
+declare const constants_COST: typeof COST;
 declare const constants_DEFAULT: typeof DEFAULT;
 declare const constants_EFFORT: typeof EFFORT;
 type constants_LlmEffort = LlmEffort;
+type constants_LlmModelCost = LlmModelCost;
 type constants_LlmProviderName = LlmProviderName;
 declare const constants_MODEL: typeof MODEL;
 declare const constants_PROVIDER: typeof PROVIDER;
 declare namespace constants {
-  export { constants_ALL as ALL, constants_DEFAULT as DEFAULT, constants_EFFORT as EFFORT, constants_MODEL as MODEL, constants_PROVIDER as PROVIDER };
-  export type { constants_LlmEffort as LlmEffort, constants_LlmProviderName as LlmProviderName };
+  export { constants_ALL as ALL, constants_COST as COST, constants_DEFAULT as DEFAULT, constants_EFFORT as EFFORT, constants_MODEL as MODEL, constants_PROVIDER as PROVIDER };
+  export type { constants_LlmEffort as LlmEffort, constants_LlmModelCost as LlmModelCost, constants_LlmProviderName as LlmProviderName };
 }
 
 interface LlmTool {
@@ -1042,4 +1108,4 @@ declare class XaiProvider implements LlmProvider {
 }
 
 export { BedrockProvider, ErrorCategory, FireworksProvider, GoogleProvider as GeminiProvider, GoogleProvider, JaypieToolkit, constants as LLM, Llm, LlmError, LlmMessageRole, LlmMessageType, LlmProgressEventType, LlmQuotaError, LlmRateLimitError, LlmStreamChunkType, LlmTransientError, LlmUnrecoverableError, OpenRouterProvider, Toolkit, XaiProvider, extractReasoning, isLlmOperateInput, isLlmOperateInputContent, isLlmOperateInputFile, isLlmOperateInputImage, jsonSchemaToNaturalSchema, naturalSchemaToJsonSchema, toolkit, tools };
-export type { LlmCache, LlmEffort, LlmErrorOptions, LlmExchangeCallback, LlmExchangeEnvelope, LlmExchangeRequest, LlmExchangeResolution, LlmExchangeResponse, LlmExchangeTiming, LlmFallbackConfig, LlmHistory, LlmInputContent, LlmInputContentFile, LlmInputContentImage, LlmInputContentText, LlmInputMessage, LlmMessageOptions, LlmModelOption, LlmOperateInput, LlmOperateInputContent, LlmOperateInputFile, LlmOperateInputImage, LlmOperateOptions, LlmOperateResponse, LlmOptions, LlmProgressCallback, LlmProgressEvent, LlmProgressToolCall, LlmProvider, LlmStreamChunk, LlmStreamChunkDone, LlmStreamChunkError, LlmStreamChunkText, LlmStreamChunkToolCall, LlmStreamChunkToolResult, LlmTool };
+export type { LlmCache, LlmEffort, LlmErrorOptions, LlmExchangeCallback, LlmExchangeEnvelope, LlmExchangeRequest, LlmExchangeResolution, LlmExchangeResponse, LlmExchangeTiming, LlmFallbackConfig, LlmHistory, LlmInputContent, LlmInputContentFile, LlmInputContentImage, LlmInputContentText, LlmInputMessage, LlmMessageOptions, LlmModelCost, LlmModelOption, LlmOperateInput, LlmOperateInputContent, LlmOperateInputFile, LlmOperateInputImage, LlmOperateOptions, LlmOperateResponse, LlmOptions, LlmProgressCallback, LlmProgressEvent, LlmProgressToolCall, LlmProvider, LlmStreamChunk, LlmStreamChunkDone, LlmStreamChunkError, LlmStreamChunkText, LlmStreamChunkToolCall, LlmStreamChunkToolResult, LlmTool };

package/dist/cjs/index.d.ts

@@ -1,6 +1,6 @@
 export { default as Llm } from "./Llm.js";
 export * as LLM from "./constants.js";
-export type { LlmEffort } from "./constants.js";
+export type { LlmEffort, LlmModelCost } from "./constants.js";
 export type { LlmCache, LlmExchangeCallback, LlmExchangeEnvelope, LlmExchangeRequest, LlmExchangeResolution, LlmExchangeResponse, LlmExchangeTiming, LlmFallbackConfig, LlmHistory, LlmInputContent, LlmInputContentFile, LlmInputContentImage, LlmInputContentText, LlmInputMessage, LlmMessageOptions, LlmModelOption, LlmOperateInput, LlmOperateInputContent, LlmOperateInputFile, LlmOperateInputImage, LlmOperateOptions, LlmOperateResponse, LlmOptions, LlmProgressCallback, LlmProgressEvent, LlmProgressToolCall, LlmProvider, } from "./types/LlmProvider.interface.js";
 export { LlmMessageRole, LlmMessageType, LlmProgressEventType, } from "./types/LlmProvider.interface.js";
 export { isLlmOperateInput, isLlmOperateInputContent, isLlmOperateInputFile, isLlmOperateInputImage, } from "./types/LlmOperateInput.guards.js";

package/dist/cjs/operate/adapters/BedrockAdapter.d.ts

@@ -47,9 +47,11 @@ type BedrockRequest = Omit<ConverseCommandInput, "messages"> & {
 type AnnotatedBedrockResponse = ConverseCommandOutput & {
     __jaypieStructuredOutput?: boolean;
 };
+/** Exported for tests; not part of the package's public surface. */
+export declare function isCachePointUnsupportedError(error: unknown): boolean;
 export declare class BedrockAdapter extends BaseProviderAdapter {
     readonly name: "bedrock";
-    readonly defaultModel: "amazon.nova-pro-v1:0";
+    readonly defaultModel: string;
     private _modelsFallbackToStructuredOutputTool;
     private _modelsWithoutTemperature;
     private _modelsWithoutCachePoint;

package/dist/esm/constants.d.ts

@@ -16,6 +16,8 @@ export declare const EFFORT: {
 };
 export type LlmEffort = (typeof EFFORT)[keyof typeof EFFORT];
 export declare const MODEL: {
+    NOVA_LITE: string;
+    NOVA_PRO: string;
     FABLE: string;
     OPUS: string;
     SONNET: string;
@@ -24,6 +26,7 @@ export declare const MODEL: {
     FIREWORKS: {
         DEEPSEEK: string;
         GLM: string;
+        GPT_OSS: string;
         KIMI: string;
         MINIMAX: string;
         QWEN: string;
@@ -47,17 +50,78 @@ export declare const MODEL: {
         SONNET: string;
     };
 };
+/** Price of one million tokens, in US dollars. */
+export interface LlmModelCost {
+    /**
+     * Cache-read (cached input) tokens. Omitted when the provider does not price
+     * cache reads separately from uncached input.
+     */
+    cachedInputRead?: number;
+    /**
+     * Cache-write tokens. A scalar when the rate does not vary by TTL (`0` where
+     * the provider publishes a free write); keyed by the same `"5m"` / `"1h"`
+     * literals as `LlmCache` when it does, so a cost calculation reads the TTL
+     * straight off `OperateRequest.cache`. Omitted means cache writes bill at the
+     * `input` rate.
+     */
+    cachedInputWrite?: number | {
+        "1h": number;
+        "5m": number;
+    };
+    /** Uncached input tokens. */
+    input: number;
+    /** Output tokens. */
+    output: number;
+    /**
+     * Reasoning tokens, when billed at a rate other than `output`. Omitted means
+     * reasoning bills as output, which is true of every provider listed here.
+     */
+    reasoning?: number;
+}
+/**
+ * Standard list price per million tokens, keyed by literal model id (verified
+ * 2026-07-21). Keys are string literals rather than `MODEL.*` references so a
+ * model retired from the catalog keeps its price here: historic ids stay
+ * priceable after they leave `MODEL.*`, which is what makes replaying old
+ * usage records possible.
+ *
+ * Caveats the numbers cannot carry:
+ * - **Standard rate only.** Introductory, batch, flex, priority, fast-mode, and
+ *   data-residency rates are excluded. Sonnet 5 is listed at its standard
+ *   $3/$15, not the introductory rate.
+ * - **Cache writes are Anthropic-only.** Bedrock publishes a literal $0 write
+ *   for Amazon's own models, recorded here as `cachedInputWrite: 0`. Fireworks
+ *   writes bill at the input rate. OpenAI and xAI discount reads automatically
+ *   and publish no write premium. Google charges nothing to write an implicit
+ *   cache; explicit caching bills storage per hour, a unit this table does not
+ *   carry (and one `@jaypie/llm` does not wire).
+ * - **Short-context tier.** Long-prompt surcharges are not modeled: Gemini 3.1
+ *   Pro doubles above 200K, Grok doubles at 200K, GPT-5.5 is 2x in / 1.5x out
+ *   above 272K.
+ * - **Text rates.** Gemini prices audio input higher than text/image/video.
+ * - **Proxy routes are deliberately absent.** `MODEL.OPENROUTER.*`, and any
+ *   uncatalogued Bedrock id reselling a third-party model, are another vendor's
+ *   model behind a gateway, priced per route and per region, so no single rate
+ *   here would be correct. Price those against the backend model, or against
+ *   the gateway's own published rate. Amazon's own Nova models are first-class,
+ *   not proxies, and are priced below at the standard US on-demand rate (`us.`
+ *   geo profile for Nova 2 Lite; the `global.` profile is cheaper and is not
+ *   modeled).
+ *
+ * Unlisted ids return `undefined` — callers must handle a miss.
+ */
+export declare const COST: Record<string, LlmModelCost>;
 export declare const PROVIDER: {
     readonly BEDROCK: {
-        readonly DEFAULT: "amazon.nova-pro-v1:0";
-        /** @deprecated Size tiers are retired in 2.0. Use PROVIDER.BEDROCK.DEFAULT. */
+        readonly DEFAULT: string;
+        /** @deprecated Size tiers are retired in 2.0. Use PROVIDER.BEDROCK.DEFAULT, or pick a specific model from MODEL.NOVA_*. */
         readonly MODEL: {
-            readonly DEFAULT: "amazon.nova-lite-v1:0";
-            readonly LARGE: "amazon.nova-pro-v1:0";
-            readonly SMALL: "amazon.nova-lite-v1:0";
-            readonly TINY: "amazon.nova-micro-v1:0";
+            readonly DEFAULT: string;
+            readonly LARGE: string;
+            readonly SMALL: string;
+            readonly TINY: string;
         };
-        readonly MODEL_MATCH_WORDS: readonly ["amazon.nova", "amazon.titan", "anthropic.claude", "cohere.command", "meta.llama", "mistral.mistral", "ai21."];
+        readonly MODEL_MATCH_WORDS: readonly ["ai21.", "amazon.nova", "amazon.titan", "anthropic.claude", "cohere.command", "deepseek.", "google.gemma", "meta.llama", "mistral.mistral", "moonshotai.", "openai.gpt-oss", "qwen."];
         readonly NAME: "bedrock";
         readonly REGION: "AWS_REGION";
     };
@@ -201,7 +265,7 @@ export declare const DEFAULT: {
  */
 export declare const ALL: {
     readonly BASE: readonly ["claude-sonnet-4-6", "gemini-3.1-pro-preview", "gpt-5.4", "grok-latest"];
-    readonly COMBINED: readonly ("claude-opus-4-8" | "claude-haiku-4-5" | "gemini-3.5-flash" | "gemini-3.1-flash-lite" | "gemini-3.1-pro-preview" | "gpt-5.5" | "gpt-5.4-mini" | "gpt-5.4-nano" | "grok-latest" | "claude-sonnet-4-6" | "gpt-5.4" | "grok-4-1-fast-reasoning" | "grok-4-1-fast-non-reasoning")[];
+    readonly COMBINED: readonly ("claude-opus-4-8" | "claude-haiku-4-5" | "gemini-3.1-pro-preview" | "gpt-5.5" | "gpt-5.4-mini" | "gpt-5.4-nano" | "grok-latest" | "claude-sonnet-4-6" | "gemini-3.1-flash-lite" | "gemini-3.5-flash" | "gpt-5.4" | "grok-4-1-fast-non-reasoning" | "grok-4-1-fast-reasoning")[];
     readonly LARGE: readonly ["claude-opus-4-8", "gemini-3.1-pro-preview", "gpt-5.5", "grok-latest"];
     readonly SMALL: readonly ["claude-sonnet-4-6", "gemini-3.5-flash", "gpt-5.4-mini", "grok-4-1-fast-reasoning"];
     readonly TINY: readonly ["claude-haiku-4-5", "gemini-3.1-flash-lite", "gpt-5.4-nano", "grok-4-1-fast-non-reasoning"];

package/dist/esm/index.d.ts

@@ -21,6 +21,8 @@ declare const EFFORT: {
 };
 type LlmEffort = (typeof EFFORT)[keyof typeof EFFORT];
 declare const MODEL: {
+    NOVA_LITE: string;
+    NOVA_PRO: string;
     FABLE: string;
     OPUS: string;
     SONNET: string;
@@ -29,6 +31,7 @@ declare const MODEL: {
     FIREWORKS: {
         DEEPSEEK: string;
         GLM: string;
+        GPT_OSS: string;
         KIMI: string;
         MINIMAX: string;
         QWEN: string;
@@ -52,17 +55,78 @@ declare const MODEL: {
         SONNET: string;
     };
 };
+/** Price of one million tokens, in US dollars. */
+interface LlmModelCost {
+    /**
+     * Cache-read (cached input) tokens. Omitted when the provider does not price
+     * cache reads separately from uncached input.
+     */
+    cachedInputRead?: number;
+    /**
+     * Cache-write tokens. A scalar when the rate does not vary by TTL (`0` where
+     * the provider publishes a free write); keyed by the same `"5m"` / `"1h"`
+     * literals as `LlmCache` when it does, so a cost calculation reads the TTL
+     * straight off `OperateRequest.cache`. Omitted means cache writes bill at the
+     * `input` rate.
+     */
+    cachedInputWrite?: number | {
+        "1h": number;
+        "5m": number;
+    };
+    /** Uncached input tokens. */
+    input: number;
+    /** Output tokens. */
+    output: number;
+    /**
+     * Reasoning tokens, when billed at a rate other than `output`. Omitted means
+     * reasoning bills as output, which is true of every provider listed here.
+     */
+    reasoning?: number;
+}
+/**
+ * Standard list price per million tokens, keyed by literal model id (verified
+ * 2026-07-21). Keys are string literals rather than `MODEL.*` references so a
+ * model retired from the catalog keeps its price here: historic ids stay
+ * priceable after they leave `MODEL.*`, which is what makes replaying old
+ * usage records possible.
+ *
+ * Caveats the numbers cannot carry:
+ * - **Standard rate only.** Introductory, batch, flex, priority, fast-mode, and
+ *   data-residency rates are excluded. Sonnet 5 is listed at its standard
+ *   $3/$15, not the introductory rate.
+ * - **Cache writes are Anthropic-only.** Bedrock publishes a literal $0 write
+ *   for Amazon's own models, recorded here as `cachedInputWrite: 0`. Fireworks
+ *   writes bill at the input rate. OpenAI and xAI discount reads automatically
+ *   and publish no write premium. Google charges nothing to write an implicit
+ *   cache; explicit caching bills storage per hour, a unit this table does not
+ *   carry (and one `@jaypie/llm` does not wire).
+ * - **Short-context tier.** Long-prompt surcharges are not modeled: Gemini 3.1
+ *   Pro doubles above 200K, Grok doubles at 200K, GPT-5.5 is 2x in / 1.5x out
+ *   above 272K.
+ * - **Text rates.** Gemini prices audio input higher than text/image/video.
+ * - **Proxy routes are deliberately absent.** `MODEL.OPENROUTER.*`, and any
+ *   uncatalogued Bedrock id reselling a third-party model, are another vendor's
+ *   model behind a gateway, priced per route and per region, so no single rate
+ *   here would be correct. Price those against the backend model, or against
+ *   the gateway's own published rate. Amazon's own Nova models are first-class,
+ *   not proxies, and are priced below at the standard US on-demand rate (`us.`
+ *   geo profile for Nova 2 Lite; the `global.` profile is cheaper and is not
+ *   modeled).
+ *
+ * Unlisted ids return `undefined` — callers must handle a miss.
+ */
+declare const COST: Record<string, LlmModelCost>;
 declare const PROVIDER: {
     readonly BEDROCK: {
-        readonly DEFAULT: "amazon.nova-pro-v1:0";
-        /** @deprecated Size tiers are retired in 2.0. Use PROVIDER.BEDROCK.DEFAULT. */
+        readonly DEFAULT: string;
+        /** @deprecated Size tiers are retired in 2.0. Use PROVIDER.BEDROCK.DEFAULT, or pick a specific model from MODEL.NOVA_*. */
         readonly MODEL: {
-            readonly DEFAULT: "amazon.nova-lite-v1:0";
-            readonly LARGE: "amazon.nova-pro-v1:0";
-            readonly SMALL: "amazon.nova-lite-v1:0";
-            readonly TINY: "amazon.nova-micro-v1:0";
+            readonly DEFAULT: string;
+            readonly LARGE: string;
+            readonly SMALL: string;
+            readonly TINY: string;
         };
-        readonly MODEL_MATCH_WORDS: readonly ["amazon.nova", "amazon.titan", "anthropic.claude", "cohere.command", "meta.llama", "mistral.mistral", "ai21."];
+        readonly MODEL_MATCH_WORDS: readonly ["ai21.", "amazon.nova", "amazon.titan", "anthropic.claude", "cohere.command", "deepseek.", "google.gemma", "meta.llama", "mistral.mistral", "moonshotai.", "openai.gpt-oss", "qwen."];
         readonly NAME: "bedrock";
         readonly REGION: "AWS_REGION";
     };
@@ -206,22 +270,24 @@ declare const DEFAULT: {
  */
 declare const ALL: {
     readonly BASE: readonly ["claude-sonnet-4-6", "gemini-3.1-pro-preview", "gpt-5.4", "grok-latest"];
-    readonly COMBINED: readonly ("claude-opus-4-8" | "claude-haiku-4-5" | "gemini-3.5-flash" | "gemini-3.1-flash-lite" | "gemini-3.1-pro-preview" | "gpt-5.5" | "gpt-5.4-mini" | "gpt-5.4-nano" | "grok-latest" | "claude-sonnet-4-6" | "gpt-5.4" | "grok-4-1-fast-reasoning" | "grok-4-1-fast-non-reasoning")[];
+    readonly COMBINED: readonly ("claude-opus-4-8" | "claude-haiku-4-5" | "gemini-3.1-pro-preview" | "gpt-5.5" | "gpt-5.4-mini" | "gpt-5.4-nano" | "grok-latest" | "claude-sonnet-4-6" | "gemini-3.5-flash" | "gemini-3.1-flash-lite" | "gpt-5.4" | "grok-4-1-fast-reasoning" | "grok-4-1-fast-non-reasoning")[];
     readonly LARGE: readonly ["claude-opus-4-8", "gemini-3.1-pro-preview", "gpt-5.5", "grok-latest"];
     readonly SMALL: readonly ["claude-sonnet-4-6", "gemini-3.5-flash", "gpt-5.4-mini", "grok-4-1-fast-reasoning"];
     readonly TINY: readonly ["claude-haiku-4-5", "gemini-3.1-flash-lite", "gpt-5.4-nano", "grok-4-1-fast-non-reasoning"];
 };
 
 declare const constants_ALL: typeof ALL;
+declare const constants_COST: typeof COST;
 declare const constants_DEFAULT: typeof DEFAULT;
 declare const constants_EFFORT: typeof EFFORT;
 type constants_LlmEffort = LlmEffort;
+type constants_LlmModelCost = LlmModelCost;
 type constants_LlmProviderName = LlmProviderName;
 declare const constants_MODEL: typeof MODEL;
 declare const constants_PROVIDER: typeof PROVIDER;
 declare namespace constants {
-  export { constants_ALL as ALL, constants_DEFAULT as DEFAULT, constants_EFFORT as EFFORT, constants_MODEL as MODEL, constants_PROVIDER as PROVIDER };
-  export type { constants_LlmEffort as LlmEffort, constants_LlmProviderName as LlmProviderName };
+  export { constants_ALL as ALL, constants_COST as COST, constants_DEFAULT as DEFAULT, constants_EFFORT as EFFORT, constants_MODEL as MODEL, constants_PROVIDER as PROVIDER };
+  export type { constants_LlmEffort as LlmEffort, constants_LlmModelCost as LlmModelCost, constants_LlmProviderName as LlmProviderName };
 }
 
 interface LlmTool {
@@ -1042,4 +1108,4 @@ declare class XaiProvider implements LlmProvider {
 }
 
 export { BedrockProvider, ErrorCategory, FireworksProvider, GoogleProvider as GeminiProvider, GoogleProvider, JaypieToolkit, constants as LLM, Llm, LlmError, LlmMessageRole, LlmMessageType, LlmProgressEventType, LlmQuotaError, LlmRateLimitError, LlmStreamChunkType, LlmTransientError, LlmUnrecoverableError, OpenRouterProvider, Toolkit, XaiProvider, extractReasoning, isLlmOperateInput, isLlmOperateInputContent, isLlmOperateInputFile, isLlmOperateInputImage, jsonSchemaToNaturalSchema, naturalSchemaToJsonSchema, toolkit, tools };
-export type { LlmCache, LlmEffort, LlmErrorOptions, LlmExchangeCallback, LlmExchangeEnvelope, LlmExchangeRequest, LlmExchangeResolution, LlmExchangeResponse, LlmExchangeTiming, LlmFallbackConfig, LlmHistory, LlmInputContent, LlmInputContentFile, LlmInputContentImage, LlmInputContentText, LlmInputMessage, LlmMessageOptions, LlmModelOption, LlmOperateInput, LlmOperateInputContent, LlmOperateInputFile, LlmOperateInputImage, LlmOperateOptions, LlmOperateResponse, LlmOptions, LlmProgressCallback, LlmProgressEvent, LlmProgressToolCall, LlmProvider, LlmStreamChunk, LlmStreamChunkDone, LlmStreamChunkError, LlmStreamChunkText, LlmStreamChunkToolCall, LlmStreamChunkToolResult, LlmTool };
+export type { LlmCache, LlmEffort, LlmErrorOptions, LlmExchangeCallback, LlmExchangeEnvelope, LlmExchangeRequest, LlmExchangeResolution, LlmExchangeResponse, LlmExchangeTiming, LlmFallbackConfig, LlmHistory, LlmInputContent, LlmInputContentFile, LlmInputContentImage, LlmInputContentText, LlmInputMessage, LlmMessageOptions, LlmModelCost, LlmModelOption, LlmOperateInput, LlmOperateInputContent, LlmOperateInputFile, LlmOperateInputImage, LlmOperateOptions, LlmOperateResponse, LlmOptions, LlmProgressCallback, LlmProgressEvent, LlmProgressToolCall, LlmProvider, LlmStreamChunk, LlmStreamChunkDone, LlmStreamChunkError, LlmStreamChunkText, LlmStreamChunkToolCall, LlmStreamChunkToolResult, LlmTool };