@oh-my-pi/pi-catalog 18.3.4 → 18.4.0

package/CHANGELOG.md

@@ -2,6 +2,19 @@
 
 ## [Unreleased]
 
+## [18.4.0] - 2026-09-28
+
+### Fixed
+
+- Corrected Cerebras model pricing so cached input tokens are charged at the model’s input rate, consistent with Cerebras billing.
+
+## [18.3.5] - 2026-09-27
+
+### Added
+
+- Added the `openai` web-search grounding for OpenAI API models that support Responses web search (`gpt-5.5`, `gpt-5.6-luna`, `gpt-6-astra`, `gpt-6-luna`) ([#13467](https://github.com/can1357/oh-my-pi/pull/13467) by [@anatoli-tsinovoy](https://github.com/anatoli-tsinovoy)).
+- Added `Model.promptCache`, per-retention-tier prompt-cache entry lifetimes in seconds (`short` / `long`), declared per provider through the `prompt-cache` KDL rule (bundled: direct Anthropic, 5 min / 1 h). Custom models and `modelOverrides` opt in with the models.yml `promptCache` key ([#12699](https://github.com/can1357/oh-my-pi/pull/12699) by [@KamijoToma](https://github.com/KamijoToma)).
+
 ## [18.3.1] - 2026-09-25
 
 ### Added

package/dist/types/build.d.ts

@@ -9,8 +9,8 @@
 import { type Api, type Model, type ModelSpec } from "./types.js";
 /**
  * Applies reviewed catalog-data value corrections (`cost-patch`,
- * `limits-patch`, `long-context-cost`, `context-window-floor`,
- * `input-modalities`) onto an upstream-sourced spec. Applied by
+ * `cache-read-at-input-rate`, `limits-patch`, `long-context-cost`,
+ * `context-window-floor`, `input-modalities`) onto an upstream-sourced spec. Applied by
  * `buildModel` to every upstream-sourced spec; user-authored overrides are
  * recomposed after building by the override applicators, so explicit user
  * limits and pricing still win.

package/dist/types/discovery/cursor-proto.d.ts

@@ -5070,6 +5070,11 @@ export interface TruncatedToolCallSuccess extends ProtoMessage {
 export declare const TruncatedToolCallSuccessSchema: MessageCodec<TruncatedToolCallSuccess>;
 /** Cursor agent message agent.v1.TurnEndedUpdate. */
 export interface TurnEndedUpdate extends ProtoMessage {
+    inputTokens?: bigint;
+    outputTokens?: bigint;
+    cacheReadTokens?: bigint;
+    cacheWriteTokens?: bigint;
+    reasoningTokens?: bigint;
 }
 export declare const TurnEndedUpdateSchema: MessageCodec<TurnEndedUpdate>;
 /** Cursor agent message agent.v1.TypeAction. */

package/dist/types/types.d.ts

@@ -12,7 +12,7 @@ export type ModelKind = (typeof MODEL_KINDS)[number];
 export declare const KIND_API_KINDS: readonly ["image", "tts", "stt", "embedding", "rerank", "video"];
 export type KindApiKind = (typeof KIND_API_KINDS)[number];
 /** Grounding transport available to chat models selected by the web role. */
-export type WebSearchGrounding = "gemini" | "anthropic" | "codex" | "xai" | "openrouter";
+export type WebSearchGrounding = "gemini" | "anthropic" | "codex" | "xai" | "openrouter" | "openai";
 /** Non-chat runner protocols accepted by catalog seeds, outside the chat dispatch union. */
 export declare const RUNNER_APIS: readonly ["local-inference", "web-search", "typesafe", "openrouter-decisions", "openai-images", "openrouter-images", "xai-tts", "openai-speech", "openai-embeddings", "openrouter-rerank", "openrouter-video", "openai-transcriptions"];
 /** Resolve a model's kind while preserving chat semantics for existing catalog rows. */
@@ -951,6 +951,12 @@ export interface TimeBasedCost {
     peakWindows: readonly PeakPricingWindow[];
     effectiveRates?: readonly EffectiveTokenCost[];
 }
+/**
+ * Best-effort prompt-cache entry lifetime in seconds for each retention tier a
+ * request can ask for. A missing tier means the lifetime is unknown; consumers
+ * must not schedule cache warming against unknown lifetimes.
+ */
+export type ModelPromptCache = Partial<Record<"short" | "long", number>>;
 /** Base token rates plus optional long-context and time-based pricing. */
 export interface ModelCost extends TokenCost {
     longContext?: LongContextTokenCost;
@@ -1077,6 +1083,14 @@ export interface Model<TApi extends Api = Api> {
      */
     accountAccess?: Readonly<Record<string, ModelAccountAccess>>;
     cost: ModelCost;
+    /**
+     * Prompt-cache entry lifetime per retention tier, in seconds. Populated only
+     * for providers whose cache-expiry and replay behavior has been validated
+     * for cache warming (direct Anthropic: 5m / 1h). A missing entry means the
+     * provider's cache behavior is unknown — such models are never warmed.
+     * Custom models and provider overrides opt in via models.yml `promptCache`.
+     */
+    promptCache?: ModelPromptCache;
     /** Premium Copilot requests charged per user-initiated request (defaults to 1). */
     premiumMultiplier?: number;
     contextWindow: number | null;

package/package.json

@@ -1,6 +1,6 @@
 {
 	"name": "@oh-my-pi/pi-catalog",
-	"version": "18.3.4",
+	"version": "18.4.0",
 	"description": "Model catalog for omp: bundled model database, provider discovery descriptors, model identity, classification, and equivalence",
 	"keywords": [
 		"ai",
@@ -93,12 +93,12 @@
 		"gen:proto": "bun scripts/generate-protocols.ts"
 	},
 	"dependencies": {
-		"@oh-my-pi/omptype": "18.3.4",
-		"@oh-my-pi/pi-utils": "18.3.4"
+		"@oh-my-pi/omptype": "18.4.0",
+		"@oh-my-pi/pi-utils": "18.4.0"
 	},
 	"devDependencies": {
 		"@bgotink/kdl": "0.4.0",
-		"@oh-my-pi/pi-ai": "18.3.4",
+		"@oh-my-pi/pi-ai": "18.4.0",
 		"@types/bun": "^1.3.14"
 	},
 	"engines": {

package/src/build.ts

@@ -82,7 +82,8 @@ function applyCatalogAssignments<TApi extends Api>(model: Model<TApi>, catalog:
 		webSearch === "anthropic" ||
 		webSearch === "codex" ||
 		webSearch === "xai" ||
-		webSearch === "openrouter"
+		webSearch === "openrouter" ||
+		webSearch === "openai"
 	) {
 		model.webSearch = webSearch;
 	}
@@ -97,6 +98,15 @@ function applyCatalogAssignments<TApi extends Api>(model: Model<TApi>, catalog:
 	}
 	const priority = catalog.priority;
 	if (typeof priority === "number") model.priority = priority;
+	const promptCache = objectPayload(catalog.promptCache);
+	if (promptCache !== undefined) {
+		const short = numberField(promptCache, "short");
+		const long = numberField(promptCache, "long");
+		model.promptCache = {
+			...(short !== undefined && { short }),
+			...(long !== undefined && { long }),
+		};
+	}
 	const applyPatchToolType = catalog.applyPatchToolType;
 	if (applyPatchToolType === "freeform" || applyPatchToolType === "function") {
 		model.applyPatchToolType = applyPatchToolType;
@@ -130,8 +140,8 @@ function applyCatalogAssignments<TApi extends Api>(model: Model<TApi>, catalog:
 
 /**
  * Applies reviewed catalog-data value corrections (`cost-patch`,
- * `limits-patch`, `long-context-cost`, `context-window-floor`,
- * `input-modalities`) onto an upstream-sourced spec. Applied by
+ * `cache-read-at-input-rate`, `limits-patch`, `long-context-cost`,
+ * `context-window-floor`, `input-modalities`) onto an upstream-sourced spec. Applied by
  * `buildModel` to every upstream-sourced spec; user-authored overrides are
  * recomposed after building by the override applicators, so explicit user
  * limits and pricing still win.
@@ -210,6 +220,14 @@ export function applyCatalogCorrections(
 			applyEffectiveFallbackRates(model.cost, Reflect.get(fallback, "effectiveRates"));
 		}
 	}
+	if (catalog.cacheReadAtInputRate === true) {
+		const { longContext } = model.cost;
+		model.cost = {
+			...model.cost,
+			cacheRead: model.cost.input,
+			...(longContext && { longContext: { ...longContext, cacheRead: longContext.input } }),
+		};
+	}
 	if (catalog.timeBased !== undefined) {
 		model.cost = { ...model.cost, timeBased: materializeTimeBasedCost(catalog.timeBased) };
 	}

package/src/compat/axes.ts

@@ -321,6 +321,16 @@ export const AXES: Readonly<Record<string, AxisDef>> = {
 	"context-window-floor": { key: "contextWindowFloor", set: "catalog", shape: "scalar" },
 	"cost-patch": { key: "costPatch", set: "catalog", shape: "object" },
 	"cost-fallback": { key: "costFallback", set: "catalog", shape: "object" },
+	/**
+	 * The host bills cache-hit input tokens at the full input rate (no cache
+	 * discount), so the built row's `cacheRead` tracks its live `input` price.
+	 */
+	"cache-read-at-input-rate": {
+		key: "cacheReadAtInputRate",
+		set: "catalog",
+		shape: "scalar",
+		values: [true, false],
+	},
 	"delegation-bias": { key: "delegationBias", set: "catalog", shape: "scalar", values: DELEGATION_BIASES },
 	"discovery-api": { key: "discoveryApi", set: "catalog", shape: "scalar" },
 	"edit-prompt-variant": { key: "editPromptVariant", set: "catalog", shape: "scalar", values: ["full", "compact"] },
@@ -331,10 +341,11 @@ export const AXES: Readonly<Record<string, AxisDef>> = {
 		key: "webSearch",
 		set: "catalog",
 		shape: "scalar",
-		values: ["gemini", "anthropic", "codex", "xai", "openrouter"],
+		values: ["gemini", "anthropic", "codex", "xai", "openrouter", "openai"],
 	},
 	"limits-patch": { key: "limitsPatch", set: "catalog", shape: "object" },
 	"long-context-cost": { key: "longContext", set: "catalog", shape: "object" },
+	"prompt-cache": { key: "promptCache", set: "catalog", shape: "object" },
 	"long-usage-limit-fallback": { key: "longUsageLimitFallback", set: "catalog", shape: "scalar" },
 	"max-context-window": { key: "maxContextWindow", set: "catalog", shape: "scalar" },
 	"requires-cursor-tool-schema-projection": {

package/src/compat/rules/providers/anthropic.kdl

@@ -6,6 +6,14 @@ provider "anthropic" {
 	discovery label="Anthropic"
 	web-search "anthropic"
 
+	// Prompt-cache entry lifetimes per retention tier, in seconds; feeds the
+	// coding agent's cache warmer. 5m writes for pay-per-token keys, 1h for
+	// retention-tier "long" requests.
+	prompt-cache {
+		short 300
+		long 3600
+	}
+
 	// Curated Anthropic models that are live or limited-availability on the
 	// first-party `/v1/models` endpoint but that stencil.so has not catalogued
 	// yet. Seeded so the bundled catalog is never gated on stencil.so's update

package/src/compat/rules/providers/cerebras.kdl

@@ -10,6 +10,9 @@ provider "cerebras" {
 	// Replaces the Cerebras provider branches in the completions baseline.
 	supports-usage-in-streaming #false
 	tool-strict-mode "all_strict"
+	// Prompt caching carries no discount: cached input tokens bill at the
+	// standard input rate (https://inference-docs.cerebras.ai/capabilities/prompt-caching).
+	cache-read-at-input-rate #true
 	// Cerebras serves Gemma 4 with vision enabled; its /v1/models rows carry
 	// no modality metadata.
 	models "gemma-4-31b" {

package/src/compat/rules/providers/openai.kdl

@@ -3,6 +3,12 @@
 provider "openai" {
 	default-model "gpt-5.5"
 	env "OPENAI_API_KEY"
+	// Responses web search is documented for GPT-5.5 and GPT-6 Astra, and has
+	// been live-verified for GPT-6 Luna and GPT-5.6 Luna. Keep this capability
+	// model-scoped so realtime and other unverified OpenAI models stay excluded.
+	models "gpt-5.5" "gpt-6-astra" "gpt-6-luna" "gpt-5.6-luna" {
+		web-search "openai"
+	}
 	kind-apis {
 		embedding "openai-embeddings"
 		image "openai-responses"