@oh-my-pi/pi-catalog 18.8.4 → 18.8.6

package/CHANGELOG.md

@@ -2,6 +2,27 @@
 
 ## [Unreleased]
 
+## [18.8.6] - 2026-10-08
+
+### Added
+
+- Added prompt-cache lookback support for Claude models across all hosts, including the public `prompt-cache-lookback` catalog axis and `resolvePromptCacheLookback` API.
+
+### Fixed
+
+- Fixed model discovery when providers publish models before they are recognized by the catalog; unsupported models are now skipped with a warning so other available models remain discoverable.
+
+## [18.8.5] - 2026-10-08
+
+### Breaking Changes
+
+- `googleAntigravityModelManagerOptions` takes `resolveAccounts` instead of `oauthToken`, and `fetchAntigravityDiscoveryModels` returns a roster or credential-rejection result instead of a bare list ([#14924](https://github.com/can1357/oh-my-pi/issues/14924)).
+
+### Fixed
+
+- Fixed Claude Haiku 5.5 on Cursor showing as unpriced; it now uses Cursor's $0.10/$0.50 rate card and 5x long-context tier above 100K input tokens ([#14890](https://github.com/can1357/oh-my-pi/pull/14890) by [@eggpeat](https://github.com/eggpeat)).
+- Fixed Claude Haiku 5.5 on GitHub Copilot pricing cache reads and writes at $0 on both the standard and `-1m` long-context rows, and the `-1m` row billing the 5x long-context band a second time above 100K input tokens ([#14890](https://github.com/can1357/oh-my-pi/pull/14890) by [@eggpeat](https://github.com/eggpeat)).
+
 ## [18.8.4] - 2026-10-08
 
 ### Fixed

package/dist/types/compat/prompt-cache-lookback.d.ts

@@ -0,0 +1,7 @@
+import type { Model } from "../types.js";
+/**
+ * Block positions the provider's prompt cache checks back from a breakpoint for
+ * an earlier request's entry (`prompt-cache-lookback` axis), or `undefined` when
+ * the model has no known lookback bound.
+ */
+export declare function resolvePromptCacheLookback(model: Model): number | undefined;

package/dist/types/discovery/antigravity.d.ts

@@ -63,13 +63,23 @@ export interface FetchAntigravityDiscoveryModelsOptions {
      */
     collapseTable?: VariantCollapseTable;
 }
+/**
+ * A complete account roster, or a definitive 401/403 credential rejection.
+ * `null` from {@link fetchAntigravityDiscoveryModels} instead means a transient
+ * failure or invalid response, which must not replace an authoritative catalog.
+ */
+export interface AntigravityModelDiscoveryResult {
+    models: ModelSpec<"google-gemini-cli">[];
+    rejectedStatus?: 401 | 403;
+}
 /**
  * Fetches discoverable Antigravity models and normalizes them into canonical model entries.
  *
- * Returns `null` on network/payload/auth failures.
- * Returns `[]` only when the endpoint responds successfully with no usable models.
+ * Returns `null` on network, server, or payload failures; a 401/403 is reported
+ * separately so multi-account discovery can skip a rejected credential.
+ * A successful empty roster has `models: []` without `rejectedStatus`.
  */
-export declare function fetchAntigravityDiscoveryModels(options: FetchAntigravityDiscoveryModelsOptions): Promise<ModelSpec<"google-gemini-cli">[] | null>;
+export declare function fetchAntigravityDiscoveryModels(options: FetchAntigravityDiscoveryModelsOptions): Promise<AntigravityModelDiscoveryResult | null>;
 /** Advertised image model and serving endpoint for one Antigravity account. */
 export interface AntigravityImageModel {
     id: string;

package/dist/types/provider-models/account-access.d.ts

@@ -0,0 +1,11 @@
+import type { Api, ModelSpec } from "../types.js";
+/**
+ * Merge per-account discovery catalogs into one list, deduped by model id.
+ *
+ * The first catalog to list an id supplies its spec; {@link ModelSpec.accountAccess}
+ * is merged from every catalog listing that id, so credential selection can
+ * route account-gated models to the accounts that serve them. Used by
+ * multi-account discovery for Codex and Antigravity, whose rosters differ per
+ * account.
+ */
+export declare function unionAccountCatalogs<TApi extends Api>(catalogs: readonly (readonly ModelSpec<TApi>[])[]): ModelSpec<TApi>[];

package/dist/types/provider-models/google.d.ts

@@ -11,8 +11,30 @@ export interface GoogleVertexModelManagerConfig {
     signal?: AbortSignal;
     fetch?: FetchImpl;
 }
+/** One Antigravity OAuth account whose `fetchAvailableModels` roster discovery reads. */
+export interface GoogleAntigravityAccount {
+    /** OAuth access token used for `Authorization: Bearer ...`. */
+    accessToken: string;
+    /**
+     * Credential identity recorded in {@link ModelSpec.accountAccess} for every
+     * model this account serves (see `oauthAccountKey` in `@oh-my-pi/pi-ai`).
+     * When any account lacks one, no model records per-account access.
+     */
+    accountKey?: string;
+}
 export interface GoogleAntigravityModelManagerConfig {
-    oauthToken?: string;
+    /**
+     * Resolves every configured Antigravity account at discovery time. Rosters
+     * are plan-scoped (Claude 5.5 is served to some accounts only) and
+     * authoritative, so each account's roster is fetched and the results are
+     * unioned; reading one account's roster would prune models only its
+     * siblings serve (#14924).
+     *
+     * Returns `null` to abort discovery (e.g. an account's credential failed to
+     * refresh), keeping the previous/bundled catalog instead of caching a
+     * partial one.
+     */
+    resolveAccounts?: () => Promise<readonly GoogleAntigravityAccount[] | null>;
     endpoint?: string;
     fetch?: FetchImpl;
 }

package/dist/types/types.d.ts

@@ -1209,11 +1209,11 @@ export interface Model<TApi extends Api = Api> {
     cursorMaxModeRoutes?: Readonly<Record<string, boolean>>;
     /**
      * Per-account availability recorded by multi-account discovery: provider
-     * account id (Codex: ChatGPT `chatgpt_account_id`) → that account's
-     * entitlements on this model. An account appears only when its own catalog
-     * lists the model, so credential selection can route account-gated models
-     * (e.g. `gpt-daybreak-blue-latest`) straight to eligible accounts. Absent on
-     * bundled/config rows and on single-account discovery.
+     * account key (Codex: ChatGPT `chatgpt_account_id`; Antigravity: login
+     * email) → that account's entitlements on this model. An account appears
+     * only when its own catalog lists the model, so credential selection can
+     * route account-gated models (e.g. `gpt-daybreak-blue-latest`, Antigravity
+     * Claude 5.5) straight to eligible accounts. Absent on bundled/config rows.
      */
     accountAccess?: Readonly<Record<string, ModelAccountAccess>>;
     /** Cursor `RequestedModel.parameters` for this model's default variant. */

package/package.json

@@ -1,6 +1,6 @@
 {
 	"name": "@oh-my-pi/pi-catalog",
-	"version": "18.8.4",
+	"version": "18.8.6",
 	"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": "18.8.4",
-		"@oh-my-pi/pi-utils": "18.8.4"
+		"@oh-my-pi/omptype": "18.8.6",
+		"@oh-my-pi/pi-utils": "18.8.6"
 	},
 	"devDependencies": {
 		"@bgotink/kdl": "0.4.0",
-		"@oh-my-pi/pi-ai": "18.8.4",
+		"@oh-my-pi/pi-ai": "18.8.6",
 		"@types/bun": "^1.3.14"
 	},
 	"engines": {

package/src/build.ts

@@ -176,40 +176,6 @@ export function applyCatalogCorrections(
 	model: Pick<ModelSpec<Api>, "cost" | "contextWindow" | "maxTokens" | "input">,
 	catalog: Record<string, unknown>,
 ): void {
-	const longContext = objectPayload(catalog.longContext);
-	if (longContext !== undefined) {
-		const inputThreshold = numberField(longContext, "inputThreshold");
-		const inclusive = Reflect.get(longContext, "inputThresholdInclusive") === true;
-		const multiplier = numberField(longContext, "multiplier");
-		const input = numberField(longContext, "input");
-		const output = numberField(longContext, "output");
-		const cacheRead = numberField(longContext, "cacheRead");
-		const cacheWrite = numberField(longContext, "cacheWrite");
-		const base = model.cost;
-		const hasTokenPrice = base.input !== 0 || base.output !== 0 || base.cacheRead !== 0 || base.cacheWrite !== 0;
-		if (inputThreshold !== undefined && multiplier !== undefined && hasTokenPrice) {
-			// Multiplier form: tier rates derive from the row's live list price.
-			model.cost = {
-				...base,
-				longContext: {
-					inputThreshold,
-					...(inclusive && { inputThresholdInclusive: true }),
-					input: base.input * multiplier,
-					output: base.output * multiplier,
-					cacheRead: base.cacheRead * multiplier,
-					cacheWrite: base.cacheWrite * multiplier,
-				},
-			};
-		} else if (
-			inputThreshold !== undefined &&
-			input !== undefined &&
-			output !== undefined &&
-			cacheRead !== undefined &&
-			cacheWrite !== undefined
-		) {
-			model.cost = { ...model.cost, longContext: { inputThreshold, input, output, cacheRead, cacheWrite } };
-		}
-	}
 	const patch = objectPayload(catalog.costPatch);
 	if (patch !== undefined) {
 		model.cost = { ...model.cost };
@@ -246,6 +212,41 @@ export function applyCatalogCorrections(
 			applyEffectiveFallbackRates(model.cost, Reflect.get(fallback, "effectiveRates"));
 		}
 	}
+	const longContext = objectPayload(catalog.longContext);
+	if (longContext !== undefined) {
+		const inputThreshold = numberField(longContext, "inputThreshold");
+		const inclusive = Reflect.get(longContext, "inputThresholdInclusive") === true;
+		const multiplier = numberField(longContext, "multiplier");
+		const input = numberField(longContext, "input");
+		const output = numberField(longContext, "output");
+		const cacheRead = numberField(longContext, "cacheRead");
+		const cacheWrite = numberField(longContext, "cacheWrite");
+		const base = model.cost;
+		const hasTokenPrice = base.input !== 0 || base.output !== 0 || base.cacheRead !== 0 || base.cacheWrite !== 0;
+		if (inputThreshold !== undefined && multiplier !== undefined && hasTokenPrice) {
+			// Multiplier form: tier rates derive from the row's final base card,
+			// after `cost-patch`/`cost-fallback` corrected the live list price.
+			model.cost = {
+				...base,
+				longContext: {
+					inputThreshold,
+					...(inclusive && { inputThresholdInclusive: true }),
+					input: base.input * multiplier,
+					output: base.output * multiplier,
+					cacheRead: base.cacheRead * multiplier,
+					cacheWrite: base.cacheWrite * multiplier,
+				},
+			};
+		} else if (
+			inputThreshold !== undefined &&
+			input !== undefined &&
+			output !== undefined &&
+			cacheRead !== undefined &&
+			cacheWrite !== undefined
+		) {
+			model.cost = { ...model.cost, longContext: { inputThreshold, input, output, cacheRead, cacheWrite } };
+		}
+	}
 	if (catalog.cacheReadAtInputRate === true) {
 		const { longContext } = model.cost;
 		model.cost = {

package/src/compat/axes.ts

@@ -399,6 +399,12 @@ export const AXES: Readonly<Record<string, AxisDef>> = {
 	"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" },
+	/**
+	 * Prompt-cache lookback in block positions: how far back from a cache
+	 * breakpoint the provider looks for an earlier request's cache entry.
+	 * Unassigned: no known lookback bound.
+	 */
+	"prompt-cache-lookback": { key: "promptCacheLookback", set: "catalog", shape: "scalar" },
 	"long-usage-limit-fallback": { key: "longUsageLimitFallback", set: "catalog", shape: "scalar" },
 	"max-context-window": { key: "maxContextWindow", set: "catalog", shape: "scalar" },
 	"pricing-status": {

package/src/compat/prompt-cache-lookback.ts

@@ -0,0 +1,23 @@
+import { toModelSpec } from "../provider-models/bundled-references";
+import type { Model } from "../types";
+import { resolveCatalogAxes } from "./resolve";
+
+const lookbackCache = new Map<string, number | null>();
+const LOOKBACK_CACHE_MAX = 8192;
+
+/**
+ * Block positions the provider's prompt cache checks back from a breakpoint for
+ * an earlier request's entry (`prompt-cache-lookback` axis), or `undefined` when
+ * the model has no known lookback bound.
+ */
+export function resolvePromptCacheLookback(model: Model): number | undefined {
+	const key = `${model.provider} ${model.id} ${model.api}`;
+	let lookback = lookbackCache.get(key);
+	if (lookback === undefined) {
+		const value = resolveCatalogAxes(toModelSpec(model)).promptCacheLookback;
+		lookback = typeof value === "number" && Number.isInteger(value) && value > 0 ? value : null;
+		if (lookbackCache.size >= LOOKBACK_CACHE_MAX) lookbackCache.clear();
+		lookbackCache.set(key, lookback);
+	}
+	return lookback ?? undefined;
+}

package/src/compat/rules/README.md

@@ -213,12 +213,14 @@ The three value shapes are:
 - **Object**: no arguments and a child block, including an empty block. Child names are kebab-case: an axis-directive spelling compiles to its resolved axis key (`template-reasoning-effort` → `qwenTemplateReasoningEffort`), anything else converts mechanically (`input-threshold` → `inputThreshold`); camelCase names are a compile error. `extra-body` payloads (top-level or nested) are the exception — their child names are literal wire JSON keys copied verbatim (`enable_thinking`). Each child is either one scalar or another object; arrays are not representable inside an object payload.
 
 A rule cannot assign the same resolved axis twice in one block.
-One object axis carries a computed form: `long-context-cost` accepts either the absolute rates (`input-threshold` + `input`/`output`/`cache-read`/`cache-write`) or `input-threshold` + `multiplier` (with optional `input-threshold-inclusive`), which derives the tier from the row's live base price at build time so the rule tracks upstream list-price updates (xAI's SuperGrok 200K tier). Rows without a token price carry no tier.
+One object axis carries a computed form: `long-context-cost` accepts either the absolute rates (`input-threshold` + `input`/`output`/`cache-read`/`cache-write`) or `input-threshold` + `multiplier` (with optional `input-threshold-inclusive`), which derives the tier from the row's base price at build time, after `cost-patch` and `cost-fallback`, so the rule tracks upstream list-price updates (xAI's SuperGrok 200K tier). Rows without a token price carry no tier. An empty `long-context-cost {}` adds no tier; a more specific rule uses it to opt a host out of a broader rule's band (GitHub Copilot's Haiku 5.5 rows, whose long tier is the `-1m` sibling).
 
 `context-window-authoritative #true` preserves a host's supplied context window through runtime model selection instead of applying inferred expansion or reference-price-tier caps. Explicit user context overrides still apply afterward. It applies to rows materialized through `buildModel` (discovery and regenerated bundles). See [`providers/factory-droid.kdl`](providers/factory-droid.kdl).
 
 `inline-image-byte-budget` caps retained base64 image characters for a deployment's request body. It leaves room for tool schemas and text beneath the provider's body-size limit; the coding agent reads it through `resolveInlineImageByteBudget` before sending live history. The budget applies only while `compat.officialEndpoint` holds: a custom `baseUrl` imposes its own body limit.
 
+`prompt-cache-lookback` is how many block positions back from a cache breakpoint the model's prompt cache looks for an earlier request's entry (Claude: 20, counting the breakpoint; a run of consecutive `tool_use` or `tool_result` blocks is one position). The coding agent reads it through `resolvePromptCacheLookback` and keeps warm-cache tool-result pruning inside the window; unassigned means no known bound.
+
 The routed-subscription registry axes (`upstream-rotation`, `region-upstreams-global|us|eu`, `region-limits-eu`, `credit-rates`, `list-price-from`, `routing-family`, `policy-aliases`, `entitlement`, `default-reasoning-off`) describe a gateway whose proxy fans one model out to several upstreams. They are read through `src/compat/factory-droid.ts` by Factory Droid discovery and its request provider, not materialized by `buildModel`. A provider-wide `region-upstreams-*` rule is the upstream serving table and a model rule replaces it for that region. `list-price-from "<provider>" ["<id>"]` shows a bundled row's list price beside the subscription's own billing; unlike seed values it is resolved at runtime and degrades to the seed's zero cost when the row is gone. See [`providers/factory-droid.kdl`](providers/factory-droid.kdl), whose wire and billing pool per model are `api-routes` and `quota-tiers` rules in `runtime/behavior.kdl`.
 
 ### Time-based pricing

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

@@ -9,7 +9,13 @@
 // are never undercounted; only Opus/Sonnet/Haiku revisions known to predate
 // 4.7 take the standard tier. Separator-collapsed revisions parse as two
 // digits (`claude-opus-45` is Opus 4.5 → 45), so 10–46 is standard too.
+// Claude's prompt cache finds an earlier request's entry only within 20 block
+// positions back from a breakpoint, counting the breakpoint; a run of
+// consecutive tool_use or tool_result blocks is one position. Warm-cache
+// tool-result pruning keeps its rewrites inside this window.
+// https://platform.claude.com/docs/en/build-with-claude/prompt-caching
 class "anthropic" {
+	prompt-cache-lookback 20
 	image-tokenization {
 		regime "anthropic-patch"
 		max-edge 2576
@@ -102,7 +108,8 @@ class "anthropic" {
 		}
 		// Haiku 5.5 bills every rate 5x for prompts over 100K. Derived from each
 		// host's list price so non-Anthropic hosts get the band too; the Claude
-		// API pins absolute rates further down.
+		// API and Cursor pin absolute rates, and GitHub Copilot opts out because
+		// its long tier is the `-1m` sibling row.
 		// https://platform.claude.com/docs/en/models/haiku-5-5/overview#pricing
 		revision ">=5.5 <6" {
 			long-context-cost {

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

@@ -167,6 +167,24 @@ provider "cursor" {
 			cache-write 12.5
 		}
 	}
+	// Claude Haiku 5.5: $0.10/$0.50, cache $0.01/$0.125. Requests above
+	// 100k input tokens bill every leg at 5x ($0.50/$2.50, cache
+	// $0.05/$0.625), on the 1M lane too.
+	models "claude-haiku-5-5" "claude-haiku-5-5-1m" {
+		cost-patch {
+			input 0.1
+			output 0.5
+			cache-read 0.01
+			cache-write 0.125
+		}
+		long-context-cost {
+			input-threshold 100000
+			input 0.5
+			output 2.5
+			cache-read 0.05
+			cache-write 0.625
+		}
+	}
 	// Claude 4.7 Opus: $5/$25, cache $0.50/$6.25. Up to 1M same rates.
 	models "claude-opus-4-7-high" "claude-opus-4-7-low" "claude-opus-4-7-max" "claude-opus-4-7-medium" "claude-opus-4-7-xhigh" {
 		cost-patch {

package/src/compat/rules/providers/github-copilot.kdl

@@ -76,4 +76,29 @@ provider "github-copilot" {
 		}
 		max-context-window 1050000
 	}
+	// Claude Haiku 5.5: `billing.token_prices` reports its input/output rates
+	// but no cache legs, so discovered rows priced cache reads and writes at
+	// $0. Copilot's rate card bills $0.01 read / $0.125 write up to 100K and
+	// $0.05 / $0.625 on the long tier (the `-1m` sibling).
+	// https://docs.github.com/en/copilot/reference/copilot-billing/models-and-pricing
+	// Copilot serves that tier as the `-1m` row (promotion target of the base
+	// row), so both rows opt out of the class-wide 100K 5x band: the base row
+	// would otherwise bill the band on top of the promotion, and the `-1m`
+	// row would charge 5x twice.
+	models "claude-haiku-5.5" {
+		cost-patch {
+			cache-read 0.01
+			cache-write 0.125
+		}
+		long-context-cost {
+		}
+	}
+	models "claude-haiku-5.5-1m" {
+		cost-patch {
+			cache-read 0.05
+			cache-write 0.625
+		}
+		long-context-cost {
+		}
+	}
 }