@stll/ai-catalog 0.2.2 → 0.3.0

package/dist/index.d.ts

@@ -460,6 +460,8 @@ type CapabilityOverride = {
   documentInput: boolean;
   documentInputReason: string;
   reasoningEfforts: readonly ReasoningEffort[] | null;
+  /** The most output tokens one response may carry, from the provider docs. */
+  outputTokens: number;
   temperatureSupported: boolean;
   /**
    * Every role sends tool definitions or structured output, so an
@@ -679,6 +681,70 @@ export declare const MODEL_TEMPERATURE_POLICIES: {
   readonly "mistral-medium-latest": "emit";
   readonly "mistral-small-latest": "emit";
 };
+/**
+ * The most output tokens one response of each offered model may carry
+ * (models.dev `limit.output`). Consumers must go through
+ * `getOutputTokenLimit`.
+ */
+export declare const MODEL_OUTPUT_TOKEN_LIMITS: {
+  readonly "gemini-3.8-flash": 65536;
+  readonly "gemini-3.7-flash": 65536;
+  readonly "gemini-3.6-flash": 65536;
+  readonly "gemini-3.5-flash-lite": 65536;
+  readonly "gemini-3.1-pro-preview": 65536;
+  readonly "gemini-3.5-flash": 65536;
+  readonly "gemini-3.1-flash-lite": 65536;
+  readonly "openai/gpt-6-astra": 128000;
+  readonly "openai/gpt-6-sol": 128000;
+  readonly "openai/gpt-6-luna": 128000;
+  readonly "openai/gpt-5.6-sol": 128000;
+  readonly "openai/gpt-5.6-terra": 128000;
+  readonly "openai/gpt-5.6-luna": 128000;
+  readonly "google/gemini-3.8-flash": 65536;
+  readonly "google/gemini-3.7-flash": 65536;
+  readonly "google/gemini-3.6-flash": 65536;
+  readonly "google/gemini-3.5-flash-lite": 65536;
+  readonly "google/gemini-3.1-pro-preview": 65536;
+  readonly "google/gemini-3.5-flash": 65536;
+  readonly "google/gemini-3.1-flash-lite": 65536;
+  readonly "anthropic/claude-sonnet-5": 128000;
+  readonly "anthropic/claude-opus-5": 128000;
+  readonly "anthropic/claude-opus-4.8": 128000;
+  readonly "anthropic/claude-sonnet-4.6": 128000;
+  readonly "openai/gpt-5.5": 128000;
+  readonly "openai/gpt-5.4-mini": 128000;
+  readonly "gpt-6-astra": 128000;
+  readonly "gpt-6-sol": 128000;
+  readonly "gpt-6-luna": 128000;
+  readonly "gpt-5.6": 128000;
+  readonly "gpt-5.6-terra": 128000;
+  readonly "gpt-5.6-luna": 128000;
+  readonly "gpt-5.5": 128000;
+  readonly "gpt-5.4": 128000;
+  readonly "gpt-5.4-mini": 128000;
+  readonly "gpt-5.4-nano": 128000;
+  readonly "gpt-5.2": 128000;
+  readonly "claude-sonnet-5": 128000;
+  readonly "claude-fable-5-1": 128000;
+  readonly "claude-fable-5": 128000;
+  readonly "claude-opus-5-5": 128000;
+  readonly "claude-opus-5": 128000;
+  readonly "claude-opus-4-8": 128000;
+  readonly "claude-opus-4-7": 128000;
+  readonly "claude-sonnet-4-6": 128000;
+  readonly "claude-opus-4-6": 128000;
+  readonly "claude-haiku-4-5-20251001": 64000;
+  readonly "us.anthropic.claude-sonnet-4-5-20250929-v1:0": 64000;
+  readonly "us.anthropic.claude-haiku-4-5-20251001-v1:0": 64000;
+  readonly "us.amazon.nova-pro-v1:0": 10000;
+  readonly "us.amazon.nova-lite-v1:0": 10000;
+  readonly "us.amazon.nova-micro-v1:0": 10000;
+  readonly "openai.gpt-oss-120b-1:0": 128000;
+  readonly "openai.gpt-oss-20b-1:0": 128000;
+  readonly "mistral-large-latest": 262144;
+  readonly "mistral-medium-latest": 262144;
+  readonly "mistral-small-latest": 256000;
+};
 //#endregion
 //#region src/index.d.ts
 /**
@@ -1224,6 +1290,14 @@ export declare const normalizeModelCatalogId: (modelId: string) => string;
  * a runtime string.
  */
 export declare const shouldEmitTemperature: (modelId: string) => boolean;
+/**
+ * The most output tokens one response of `modelId` may carry, from the
+ * catalog. `undefined` for an id the catalog does not list (a deployment
+ * override, a dev model): the catalog knows nothing about it, so callers leave
+ * the allowance to the provider rather than guess one. Callers must never
+ * index `MODEL_OUTPUT_TOKEN_LIMITS` directly with a runtime string.
+ */
+export declare const getOutputTokenLimit: (modelId: string) => number | undefined;
 /**
  * Whether a model accepts tool use on a streaming request.
  *

package/dist/index.js

@@ -538,6 +538,70 @@ const MODEL_TEMPERATURE_POLICIES = {
 	"mistral-medium-latest": "emit",
 	"mistral-small-latest": "emit"
 };
+/**
+* The most output tokens one response of each offered model may carry
+* (models.dev `limit.output`). Consumers must go through
+* `getOutputTokenLimit`.
+*/
+const MODEL_OUTPUT_TOKEN_LIMITS = {
+	"gemini-3.8-flash": 65536,
+	"gemini-3.7-flash": 65536,
+	"gemini-3.6-flash": 65536,
+	"gemini-3.5-flash-lite": 65536,
+	"gemini-3.1-pro-preview": 65536,
+	"gemini-3.5-flash": 65536,
+	"gemini-3.1-flash-lite": 65536,
+	"openai/gpt-6-astra": 128e3,
+	"openai/gpt-6-sol": 128e3,
+	"openai/gpt-6-luna": 128e3,
+	"openai/gpt-5.6-sol": 128e3,
+	"openai/gpt-5.6-terra": 128e3,
+	"openai/gpt-5.6-luna": 128e3,
+	"google/gemini-3.8-flash": 65536,
+	"google/gemini-3.7-flash": 65536,
+	"google/gemini-3.6-flash": 65536,
+	"google/gemini-3.5-flash-lite": 65536,
+	"google/gemini-3.1-pro-preview": 65536,
+	"google/gemini-3.5-flash": 65536,
+	"google/gemini-3.1-flash-lite": 65536,
+	"anthropic/claude-sonnet-5": 128e3,
+	"anthropic/claude-opus-5": 128e3,
+	"anthropic/claude-opus-4.8": 128e3,
+	"anthropic/claude-sonnet-4.6": 128e3,
+	"openai/gpt-5.5": 128e3,
+	"openai/gpt-5.4-mini": 128e3,
+	"gpt-6-astra": 128e3,
+	"gpt-6-sol": 128e3,
+	"gpt-6-luna": 128e3,
+	"gpt-5.6": 128e3,
+	"gpt-5.6-terra": 128e3,
+	"gpt-5.6-luna": 128e3,
+	"gpt-5.5": 128e3,
+	"gpt-5.4": 128e3,
+	"gpt-5.4-mini": 128e3,
+	"gpt-5.4-nano": 128e3,
+	"gpt-5.2": 128e3,
+	"claude-sonnet-5": 128e3,
+	"claude-fable-5-1": 128e3,
+	"claude-fable-5": 128e3,
+	"claude-opus-5-5": 128e3,
+	"claude-opus-5": 128e3,
+	"claude-opus-4-8": 128e3,
+	"claude-opus-4-7": 128e3,
+	"claude-sonnet-4-6": 128e3,
+	"claude-opus-4-6": 128e3,
+	"claude-haiku-4-5-20251001": 64e3,
+	"us.anthropic.claude-sonnet-4-5-20250929-v1:0": 64e3,
+	"us.anthropic.claude-haiku-4-5-20251001-v1:0": 64e3,
+	"us.amazon.nova-pro-v1:0": 1e4,
+	"us.amazon.nova-lite-v1:0": 1e4,
+	"us.amazon.nova-micro-v1:0": 1e4,
+	"openai.gpt-oss-120b-1:0": 128e3,
+	"openai.gpt-oss-20b-1:0": 128e3,
+	"mistral-large-latest": 262144,
+	"mistral-medium-latest": 262144,
+	"mistral-small-latest": 256e3
+};
 //#endregion
 //#region src/model-rates.gen.ts
 const MODEL_RATES = {
@@ -1625,6 +1689,15 @@ const MODEL_TEMPERATURE_POLICY_BY_ID = MODEL_TEMPERATURE_POLICIES;
 * a runtime string.
 */
 const shouldEmitTemperature = (modelId) => MODEL_TEMPERATURE_POLICY_BY_ID[normalizeModelCatalogId(modelId)] === "emit";
+const MODEL_OUTPUT_TOKEN_LIMIT_BY_ID = MODEL_OUTPUT_TOKEN_LIMITS;
+/**
+* The most output tokens one response of `modelId` may carry, from the
+* catalog. `undefined` for an id the catalog does not list (a deployment
+* override, a dev model): the catalog knows nothing about it, so callers leave
+* the allowance to the provider rather than guess one. Callers must never
+* index `MODEL_OUTPUT_TOKEN_LIMITS` directly with a runtime string.
+*/
+const getOutputTokenLimit = (modelId) => MODEL_OUTPUT_TOKEN_LIMIT_BY_ID[normalizeModelCatalogId(modelId)];
 /**
 * Whether a model accepts tool use on a streaming request.
 *
@@ -1862,4 +1935,4 @@ const DEFAULT_CONTEXT_WINDOW_TOKENS = 128e3;
 */
 const getContextWindowTokens = (modelId) => CONTEXT_WINDOW_TOKENS_BY_ID[normalizeModelCatalogId(modelId)] ?? 128e3;
 //#endregion
-export { AI_PROVIDERS, ANTHROPIC_ADAPTIVE_THINKING_MODELS, BYOK_DEFAULT_MODELS, BYOK_DOCUMENT_INPUT_MODEL_OPTIONS, BYOK_MODEL_OPTIONS, CAPABILITY_OVERRIDES, CHAT_PDF_ATTACHMENT_MODEL_OPTIONS, CONTEXT_WINDOW_TOKENS, DEFAULT_CONTEXT_WINDOW_TOKENS, DEFAULT_MODELS, DOCUMENT_INPUT_OVERRIDES, FALLBACK_CHAT_MODEL_BY_PROVIDER, FIRST_PARTY_MODEL_PROVIDERS, MODELS_DEV_RATE_PROVIDER_BY_CATALOG_PROVIDER, MODELS_DEV_RATE_SOURCE_ALIASES, MODEL_CATALOG_ID_ALIASES, MODEL_CATALOG_PROVIDER_KIND, MODEL_DEFAULT_REASONING_EFFORTS, MODEL_DISPLAY_METADATA, MODEL_RATES, MODEL_RATE_UNITS_PER_USD, MODEL_REASONING_EFFORTS, MODEL_ROLES, MODEL_STREAMING_TOOL_USE, MODEL_TEMPERATURE_POLICIES, REASONING_EFFORTS, RETAINED_MODELS_DEV_RATE_ENTRIES, STREAMING_TOOL_USE_SUPPORTS, TANSTACK_AI_PROVIDERS, TEMPERATURE_POLICIES, getContextWindowTokens, getModelDefaultReasoningEffort, getModelDisplayMetadata, getModelRate, getModelReasoningEfforts, getStandardModelRate, isBYOKModelRoleSupported, isBYOKProviderRoleSupported, isChatPdfAttachmentModelSupported, normalizeModelCatalogId, resolveModelRate, resolveReasoningEffort, resolveWorkingBYOKModelForRole, shouldEmitTemperature, supportsStreamingToolUse };
+export { AI_PROVIDERS, ANTHROPIC_ADAPTIVE_THINKING_MODELS, BYOK_DEFAULT_MODELS, BYOK_DOCUMENT_INPUT_MODEL_OPTIONS, BYOK_MODEL_OPTIONS, CAPABILITY_OVERRIDES, CHAT_PDF_ATTACHMENT_MODEL_OPTIONS, CONTEXT_WINDOW_TOKENS, DEFAULT_CONTEXT_WINDOW_TOKENS, DEFAULT_MODELS, DOCUMENT_INPUT_OVERRIDES, FALLBACK_CHAT_MODEL_BY_PROVIDER, FIRST_PARTY_MODEL_PROVIDERS, MODELS_DEV_RATE_PROVIDER_BY_CATALOG_PROVIDER, MODELS_DEV_RATE_SOURCE_ALIASES, MODEL_CATALOG_ID_ALIASES, MODEL_CATALOG_PROVIDER_KIND, MODEL_DEFAULT_REASONING_EFFORTS, MODEL_DISPLAY_METADATA, MODEL_OUTPUT_TOKEN_LIMITS, MODEL_RATES, MODEL_RATE_UNITS_PER_USD, MODEL_REASONING_EFFORTS, MODEL_ROLES, MODEL_STREAMING_TOOL_USE, MODEL_TEMPERATURE_POLICIES, REASONING_EFFORTS, RETAINED_MODELS_DEV_RATE_ENTRIES, STREAMING_TOOL_USE_SUPPORTS, TANSTACK_AI_PROVIDERS, TEMPERATURE_POLICIES, getContextWindowTokens, getModelDefaultReasoningEffort, getModelDisplayMetadata, getModelRate, getModelReasoningEfforts, getOutputTokenLimit, getStandardModelRate, isBYOKModelRoleSupported, isBYOKProviderRoleSupported, isChatPdfAttachmentModelSupported, normalizeModelCatalogId, resolveModelRate, resolveReasoningEffort, resolveWorkingBYOKModelForRole, shouldEmitTemperature, supportsStreamingToolUse };

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "@stll/ai-catalog",
-  "version": "0.2.2",
+  "version": "0.3.0",
   "description": "Provider-neutral AI model roles, capabilities, and curated model metadata.",
   "keywords": [
     "ai",