@oh-my-pi/pi-catalog 18.14.2 → 18.16.0

package/src/models.ts

@@ -1,7 +1,7 @@
 import { classifyModel } from "./compat/taxonomy";
 import { getCatalogProvider, getCatalogProviders, openModelCatalog } from "./model-catalog-format";
 import modelsData from "./models.json" with { type: "json" };
-import { dueEffectiveRate } from "./pricing";
+import { dueDatedCard, dueEffectiveRate } from "./pricing";
 import type {
 	Api,
 	EffectiveTokenCost,
@@ -77,6 +77,37 @@ export function getModelPricingStatus(model: Pick<Model<Api>, "cost" | "pricingS
 	return model.pricingStatus ?? "unknown";
 }
 
+/** Factory Droid credit rate in force at a time, and the next dated change to it. */
+export interface FactoryDroidCreditRate {
+	rate: number;
+	/** The earliest later `credit-rates` card that changes `rate`: the rate it sets and when (Unix ms). */
+	next?: { rate: number; at: number };
+}
+
+/**
+ * Factory Droid Standard Credits rate at `timestamp` (Unix ms, default now):
+ * the `input` of the `credit-rates` card due then (the latest at or before
+ * it, as for cost cards), else the base rate. Undefined for models without a
+ * credit rate.
+ */
+export function getFactoryDroidCreditRate(
+	model: Pick<Model, "factoryDroidCredits" | "factoryDroidCreditCards">,
+	timestamp = Date.now(),
+): FactoryDroidCreditRate | undefined {
+	const base = model.factoryDroidCredits;
+	if (base === undefined) return undefined;
+	const cards = model.factoryDroidCreditCards ?? [];
+	const rate = dueDatedCard(cards, timestamp)?.input ?? base;
+	let next: FactoryDroidCreditRate["next"];
+	for (const card of cards) {
+		const cardRate = card.input ?? base;
+		if (card.effectiveFrom > timestamp && cardRate !== rate && !(next && next.at <= card.effectiveFrom)) {
+			next = { rate: cardRate, at: card.effectiveFrom };
+		}
+	}
+	return next ? { rate, next } : { rate };
+}
+
 /**
  * The base token card in force at `timestamp` (Unix ms, default now): the
  * latest due `timeBased` effective rate, else the row's own legs. Price

package/src/pricing.ts

@@ -1,84 +1,35 @@
-import type {
-	EffectiveTokenCost,
-	LongContextTokenCost,
-	ModelCost,
-	PeakPricingWindow,
-	TimeBasedCost,
-	TokenCost,
-} from "./types";
+import { type } from "@oh-my-pi/omptype";
+import type { EffectiveTokenCost, ModelCost, TimeBasedCost, TokenCost } from "./types";
 import { isRecord } from "./utils";
 
-function nonnegative(value: unknown): value is number {
-	return typeof value === "number" && Number.isFinite(value) && value >= 0;
-}
-
-function isRates(value: unknown): value is TokenCost {
-	return (
-		isRecord(value) &&
-		nonnegative(value.input) &&
-		nonnegative(value.output) &&
-		nonnegative(value.cacheRead) &&
-		nonnegative(value.cacheWrite)
-	);
-}
-
-function isLongContext(value: unknown): value is LongContextTokenCost {
-	return (
-		isRecord(value) &&
-		isRates(value) &&
-		typeof value.inputThreshold === "number" &&
-		Number.isFinite(value.inputThreshold) &&
-		value.inputThreshold > 0 &&
-		(value.inputThresholdInclusive === undefined || typeof value.inputThresholdInclusive === "boolean")
-	);
-}
-
-function isPeakWindow(value: unknown): value is PeakPricingWindow {
-	return (
-		isRecord(value) &&
-		Array.isArray(value.weekdays) &&
-		value.weekdays.length > 0 &&
-		value.weekdays.every(day => Number.isInteger(day) && day >= 0 && day <= 6) &&
-		new Set(value.weekdays).size === value.weekdays.length &&
-		typeof value.startMinute === "number" &&
-		Number.isInteger(value.startMinute) &&
-		typeof value.endMinute === "number" &&
-		Number.isInteger(value.endMinute) &&
-		value.startMinute >= 0 &&
-		value.endMinute <= 1440 &&
-		value.startMinute < value.endMinute
-	);
-}
-
-function isEffectiveRate(value: unknown): value is EffectiveTokenCost {
-	return (
-		isRecord(value) &&
-		isRates(value) &&
-		typeof value.effectiveFrom === "number" &&
-		Number.isSafeInteger(value.effectiveFrom) &&
-		Math.abs(value.effectiveFrom) <= 8_640_000_000_000_000 &&
-		(value.longContext === undefined || isLongContext(value.longContext))
-	);
-}
+// Schedules are also checked on in-memory discovery output, where absent optionals may be explicit `undefined`.
+const Rate = type("number >= 0");
+const Rates = type({ input: Rate, output: Rate, cacheRead: Rate, cacheWrite: Rate });
+const LongContext = Rates.and({ inputThreshold: "number > 0", "inputThresholdInclusive?": "boolean | undefined" });
+const PeakWindow = type({
+	weekdays: type("0 <= number.integer <= 6").array().atLeastLength(1),
+	startMinute: "number.integer >= 0",
+	endMinute: "number.integer <= 1440",
+}).narrow(window => new Set(window.weekdays).size === window.weekdays.length && window.startMinute < window.endMinute);
+const EffectiveRate = Rates.and({
+	// The `Date` range.
+	effectiveFrom: "-8640000000000000 <= number.integer <= 8640000000000000",
+	"longContext?": LongContext.or("undefined"),
+});
+/** Serialized time-based pricing schedule, as cached model rows carry it. */
+export const TimeBasedCostShape = type({
+	offPeakMultiplier: Rate,
+	peakWindows: PeakWindow.array(),
+	"effectiveRates?": EffectiveRate.array().or("undefined"),
+}).narrow(
+	cost =>
+		cost.effectiveRates === undefined ||
+		new Set(cost.effectiveRates.map(rate => rate.effectiveFrom)).size === cost.effectiveRates.length,
+);
 
 /** Validate serialized schedules before admitting cached model rows. */
 export function isTimeBasedCost(value: unknown): value is TimeBasedCost {
-	if (
-		!isRecord(value) ||
-		!nonnegative(value.offPeakMultiplier) ||
-		!Array.isArray(value.peakWindows) ||
-		!value.peakWindows.every(isPeakWindow)
-	) {
-		return false;
-	}
-	if (value.effectiveRates === undefined) return true;
-	if (!Array.isArray(value.effectiveRates) || !value.effectiveRates.every(isEffectiveRate)) return false;
-	const dates = new Set<number>();
-	for (const rate of value.effectiveRates) {
-		if (dates.has(rate.effectiveFrom)) return false;
-		dates.add(rate.effectiveFrom);
-	}
-	return true;
+	return TimeBasedCostShape.allows(value);
 }
 
 function payload(value: unknown, keys: readonly string[], field: string): Record<string, unknown> {
@@ -148,15 +99,15 @@ export function materializeTimeBasedCost(value: unknown): TimeBasedCost {
 }
 
 /**
- * The latest `effectiveRates` card of a materialized schedule due at
- * `timestamp`, or undefined before the first one (or without a schedule).
+ * The dated card due at `timestamp`: the one with the latest `effectiveFrom`
+ * at or before it, in any order, or undefined before the first card.
  */
-export function dueEffectiveRate(
-	schedule: TimeBasedCost | undefined,
+export function dueDatedCard<T extends { effectiveFrom: number }>(
+	cards: readonly T[],
 	timestamp: number,
-): EffectiveTokenCost | undefined {
-	let due: EffectiveTokenCost | undefined;
-	for (const candidate of schedule?.effectiveRates ?? []) {
+): T | undefined {
+	let due: T | undefined;
+	for (const candidate of cards) {
 		if (candidate.effectiveFrom <= timestamp && (due === undefined || candidate.effectiveFrom > due.effectiveFrom)) {
 			due = candidate;
 		}
@@ -164,21 +115,38 @@ export function dueEffectiveRate(
 	return due;
 }
 
+/**
+ * The latest `effectiveRates` card of a materialized schedule due at
+ * `timestamp`, or undefined before the first one (or without a schedule).
+ */
+export function dueEffectiveRate(
+	schedule: TimeBasedCost | undefined,
+	timestamp: number,
+): EffectiveTokenCost | undefined {
+	return dueDatedCard(schedule?.effectiveRates ?? [], timestamp);
+}
+
 const TOKEN_COST_LEGS = ["input", "output", "cacheRead", "cacheWrite"] as const;
 const DATED_CARD_KEYS: readonly string[] = ["effectiveFrom", ...TOKEN_COST_LEGS];
 
-/** A `cost-patch`/`cost-fallback` dated card: the legs it overwrites from `effectiveFrom` (Unix ms). */
+/**
+ * A dated `effective-rates` card of a `cost-patch`, `cost-fallback` or
+ * `credit-rates` rule: the legs it overwrites from `effectiveFrom` (Unix ms).
+ */
 export interface DatedCostCard extends Partial<TokenCost> {
 	effectiveFrom: number;
 }
 
 /**
- * Parse the compiled `effective-rates` block of a `cost-patch` or
- * `cost-fallback` payload into dated cards, ascending by date. Throws on a
- * malformed card, so a typo in a rule fails the build instead of silently
- * leaving the old price in force.
+ * Parse the compiled `effective-rates` block of a `cost-patch`,
+ * `cost-fallback` or Factory Droid `credit-rates` payload into dated cards,
+ * ascending by date. Throws on a malformed card, so a typo in a rule fails the
+ * build instead of silently leaving the old price in force.
  */
-export function parseDatedCostCards(costAxis: unknown, axis: "cost-patch" | "cost-fallback"): DatedCostCard[] {
+export function parseDatedCostCards(
+	costAxis: unknown,
+	axis: "cost-patch" | "cost-fallback" | "credit-rates",
+): DatedCostCard[] {
 	const effectiveRates = isRecord(costAxis) ? costAxis.effectiveRates : undefined;
 	if (effectiveRates === undefined) return [];
 	if (!isRecord(effectiveRates)) throw new Error(`Invalid ${axis} effective-rates: expected named objects`);
@@ -190,7 +158,7 @@ export function parseDatedCostCards(costAxis: unknown, axis: "cost-patch" | "cos
 		for (const leg of TOKEN_COST_LEGS) {
 			const value = entry[leg];
 			if (value === undefined) continue;
-			if (!nonnegative(value))
+			if (!Rate.allows(value))
 				throw new Error(`Invalid ${axis} effective-rate ${leg}: expected a nonnegative number`);
 			card[leg] = value;
 		}
@@ -235,10 +203,7 @@ export function datedCost(base: TokenCost, cards: readonly DatedCostCard[], now
  * `cost` unchanged before the first card.
  */
 export function applyDueDatedCostCard(cost: TokenCost, cards: readonly DatedCostCard[], now = Date.now()): void {
-	let due: DatedCostCard | undefined;
-	for (const card of cards) {
-		if (card.effectiveFrom <= now) due = card;
-	}
+	const due = dueDatedCard(cards, now);
 	if (due === undefined) return;
 	for (const leg of TOKEN_COST_LEGS) {
 		const value = due[leg];

package/src/provider-models/descriptors.ts

@@ -61,6 +61,7 @@ import {
 	qwenPortalModelManagerOptions,
 	sakanaModelManagerOptions,
 	sambanovaModelManagerOptions,
+	scalewayModelManagerOptions,
 	siliconflowCnModelManagerOptions,
 	siliconflowModelManagerOptions,
 	singularityApiDevModelManagerOptions,
@@ -150,6 +151,7 @@ const MODEL_MANAGER_FACTORIES: Readonly<Partial<Record<KnownProvider, ModelManag
 	"qwen-portal": config => qwenPortalModelManagerOptions(config),
 	sakana: config => sakanaModelManagerOptions(config),
 	sambanova: config => sambanovaModelManagerOptions(config),
+	scaleway: config => scalewayModelManagerOptions(config),
 	siliconflow: config => siliconflowModelManagerOptions(config),
 	"siliconflow-cn": config => siliconflowCnModelManagerOptions(config),
 	"singularityapi-dev": config => singularityApiDevModelManagerOptions(config),

package/src/provider-models/openai-compat.ts

@@ -13,7 +13,7 @@ import {
 import { resolveAnthropicApiVersion } from "../compat/anthropic";
 import { xaiResponsesReasoningEffortMap } from "../compat/openai";
 import { hasModelScopedEffortLadder, resolveModelPolicy } from "../compat/resolve";
-import { compareRevision, parseRevision } from "../compat/revision";
+import { compareRevision, parseRevision, revisionAtLeast } from "../compat/revision";
 import { providerEntries, seedModels } from "../compat/providers";
 import { billingVariantPlain, classifyModel, discoveryVocabulary } from "../compat/taxonomy";
 import {
@@ -21,6 +21,7 @@ import {
 	fetchOpenAICompatibleModels,
 	type OpenAICompatibleModelMapperContext,
 	type OpenAICompatibleModelRecord,
+	withOpenAICompatibleDiscoveryTimeout,
 } from "../discovery/openai-compatible";
 import { Effort, THINKING_EFFORTS } from "../effort";
 import { FIREWORKS_FAST_SUFFIX, toFireworksPublicModelId } from "../fireworks-model-id";
@@ -43,15 +44,7 @@ import {
 	type TokenCost,
 	runnerApiKind,
 } from "../types";
-import {
-	discoveryFetch,
-	isAnthropicOAuthToken,
-	isRecord,
-	toBoolean,
-	toNumber,
-	toPositiveNumber,
-	toPositiveNumberOrNull,
-} from "../utils";
+import { discoveryFetch, isAnthropicOAuthToken, isRecord, toBoolean, toNumber, toPositiveNumber } from "../utils";
 import { ALIBABA_TOKEN_PLAN_BASE_URL, parseAlibabaTokenPlanCredential } from "../wire/alibaba-token-plan";
 import { normalizeCharmHyperBaseUrl } from "../wire/charm-hyper";
 import { CLINEPASS_API_BASE_URL, clinePassClientHeaders } from "../wire/cline-pass";
@@ -78,15 +71,6 @@ import { getClinePassModelMetadata } from "./cline-pass";
 import type { ModelManagerConfig } from "./descriptor-types";
 import { filterModelsDevCatalogRows } from "./models-dev-policies";
 
-function revisionAtLeast(revision: string | undefined, floor: string): boolean {
-	if (revision === undefined) return false;
-	const parsedRevision = parseRevision(revision);
-	const parsedFloor = parseRevision(floor);
-	return (
-		parsedRevision !== undefined && parsedFloor !== undefined && compareRevision(parsedRevision, parsedFloor) >= 0
-	);
-}
-
 function isGlmReasoningIdentity(provider: string, modelId: string, floor: string): boolean {
 	const identity = classifyModel(provider, modelId, { lenient: true });
 	if (identity.class !== "glm" || !revisionAtLeast(identity.revision, floor)) return false;
@@ -99,24 +83,6 @@ const MODELS_DEV_URL = "https://catalog.stencil.so/models.json.zstd";
 /** Little-endian magic number opening every zstd frame (RFC 8878). */
 const ZSTD_MAGIC = 0xfd2fb528;
 
-/**
- * Uses a cancellable timer rather than the native abort-timeout helper so
- * successful fast discovery requests do not leave armed timeout signals for
- * concurrent GC to trip over later.
- */
-async function withCatalogDiscoveryTimeout<T>(timeoutMs: number, run: (signal: AbortSignal) => Promise<T>): Promise<T> {
-	const controller = new AbortController();
-	const timer = setTimeout(
-		() => controller.abort(new DOMException("The operation timed out.", "TimeoutError")),
-		timeoutMs,
-	);
-	try {
-		return await run(controller.signal);
-	} finally {
-		clearTimeout(timer);
-	}
-}
-
 const ANTHROPIC_BASE_URL = "https://api.anthropic.com/v1";
 const ANTHROPIC_OAUTH_BETA =
 	"claude-code-20250219,oauth-2025-04-20,interleaved-thinking-2025-05-14,redact-thinking-2026-02-12,context-management-2025-06-27,prompt-caching-scope-2026-01-05,mid-conversation-system-2026-04-07,advanced-tool-use-2025-11-20,effort-2025-11-24,extended-cache-ttl-2025-04-11";
@@ -226,8 +192,9 @@ export function fetchWellKnownModels(fetchImpl?: FetchImpl, signal?: AbortSignal
 function fetchRevalidatedWellKnownModels(fetchImpl?: FetchImpl, signal?: AbortSignal): Promise<unknown> {
 	const session = getCatalogSession(fetchImpl);
 	if (!session.inflight) {
-		session.inflight = withCatalogDiscoveryTimeout(DEFAULT_OPENAI_COMPATIBLE_DISCOVERY_TIMEOUT_MS, transportSignal =>
-			fetchCatalogPayload(fetchImpl ?? discoveryFetch(), session, transportSignal),
+		session.inflight = withOpenAICompatibleDiscoveryTimeout(
+			DEFAULT_OPENAI_COMPATIBLE_DISCOVERY_TIMEOUT_MS,
+			transportSignal => fetchCatalogPayload(fetchImpl ?? discoveryFetch(), session, transportSignal),
 		).finally(() => {
 			session.inflight = null;
 		});
@@ -239,7 +206,7 @@ function fetchRevalidatedWellKnownModelsWithTimeout(
 	fetchImpl?: FetchImpl,
 	timeoutMs = DEFAULT_OPENAI_COMPATIBLE_DISCOVERY_TIMEOUT_MS,
 ): Promise<unknown> {
-	return withCatalogDiscoveryTimeout(timeoutMs, signal => fetchRevalidatedWellKnownModels(fetchImpl, signal));
+	return withOpenAICompatibleDiscoveryTimeout(timeoutMs, signal => fetchRevalidatedWellKnownModels(fetchImpl, signal));
 }
 
 async function fetchCatalogPayload(
@@ -545,6 +512,8 @@ function mapWithBundledReference<TApi extends Api>(
 		return {
 			...defaults,
 			name,
+			contextWindow: toPositiveNumber(entry.context_length, defaults.contextWindow),
+			maxTokens: toPositiveNumber(entry.max_completion_tokens, defaults.maxTokens),
 		};
 	}
 	// Generic `/models` rows describe the chat roster. Do not make a bundled
@@ -1725,8 +1694,9 @@ async function fetchDeepinfraModels(options: {
 	const fetchImpl = discoveryFetch(options.fetch);
 	let payload: unknown;
 	try {
-		const response = await withCatalogDiscoveryTimeout(DEFAULT_OPENAI_COMPATIBLE_DISCOVERY_TIMEOUT_MS, signal =>
-			fetchImpl(`${options.baseUrl}/models${DEEPINFRA_MODELS_QUERY}`, { method: "GET", headers, signal }),
+		const response = await withOpenAICompatibleDiscoveryTimeout(
+			DEFAULT_OPENAI_COMPATIBLE_DISCOVERY_TIMEOUT_MS,
+			signal => fetchImpl(`${options.baseUrl}/models${DEEPINFRA_MODELS_QUERY}`, { method: "GET", headers, signal }),
 		);
 		if (!response.ok) {
 			return null;
@@ -2284,7 +2254,7 @@ async function loadSiliconFlowModelsDevReferences(
 	try {
 		// Bounded: this enrichment is optional, so a stalled models.dev must not
 		// hold back the authoritative endpoint request that runs after it.
-		const payload = await withCatalogDiscoveryTimeout(SILICONFLOW_MODELS_DEV_REFERENCE_TIMEOUT_MS, signal =>
+		const payload = await withOpenAICompatibleDiscoveryTimeout(SILICONFLOW_MODELS_DEV_REFERENCE_TIMEOUT_MS, signal =>
 			fetchWellKnownModels(fetchImpl, signal),
 		);
 		return createModelsDevReferenceMap<"openai-completions">(
@@ -2828,7 +2798,7 @@ const CLINE_PASS_LIVE_CATALOG_URL = "https://openrouter.ai/api/v1/models";
 /** Enrichment tier, never authoritative: any failure degrades to the bundled path. */
 async function fetchClinePassLiveCatalog(fetchImpl: FetchImpl): Promise<ClinePassLiveCatalog | undefined> {
 	try {
-		const response = await withCatalogDiscoveryTimeout(5_000, signal =>
+		const response = await withOpenAICompatibleDiscoveryTimeout(5_000, signal =>
 			fetchImpl(CLINE_PASS_LIVE_CATALOG_URL, { signal }),
 		);
 		if (!response.ok) return undefined;
@@ -3007,7 +2977,7 @@ async function fetchClinePassModels(
 	// The roster is authoritative for ids; the live reference catalog is a pure
 	// enrichment tier fetched alongside and tolerated away (offline → bundle).
 	const [response, liveCatalog] = await Promise.all([
-		withCatalogDiscoveryTimeout(5_000, signal =>
+		withOpenAICompatibleDiscoveryTimeout(5_000, signal =>
 			fetchImpl(CLINEPASS_MODELS_URL, { signal, headers: clinePassClientHeaders() }),
 		),
 		fetchClinePassLiveCatalog(fetchImpl),
@@ -4422,7 +4392,7 @@ export async function fetchLmStudioNativeModelMetadata(
 	if (options?.signal !== undefined) {
 		return fetchMetadata(options.signal);
 	}
-	return withCatalogDiscoveryTimeout(LM_STUDIO_NATIVE_METADATA_TIMEOUT_MS, fetchMetadata);
+	return withOpenAICompatibleDiscoveryTimeout(LM_STUDIO_NATIVE_METADATA_TIMEOUT_MS, fetchMetadata);
 }
 
 export interface LmStudioModelManagerConfig {
@@ -4770,8 +4740,6 @@ export function basetenModelManagerOptions(
 				cacheWrite: 0,
 			};
 
-			const contextWindow = toPositiveNumber(raw.context_length, reference?.contextWindow ?? defaults.contextWindow);
-			const maxTokens = toPositiveNumber(raw.max_completion_tokens, reference?.maxTokens ?? defaults.maxTokens);
 			const baseModel = mapWithBundledReference(entry, defaults, reference);
 
 			return {
@@ -4779,8 +4747,6 @@ export function basetenModelManagerOptions(
 				reasoning,
 				input: vision ? ["text", "image"] : ["text"],
 				cost,
-				contextWindow,
-				maxTokens,
 				...(supportsTools === false ? { supportsTools } : {}),
 			};
 		},
@@ -6393,7 +6359,9 @@ async function fetchLiteLLMRichModelsInternal<TApi extends Api>(
 	if (options.signal !== undefined) {
 		return fetchModels(options.signal);
 	}
-	return options.timeoutMs !== undefined ? withCatalogDiscoveryTimeout(options.timeoutMs, fetchModels) : fetchModels();
+	return options.timeoutMs !== undefined
+		? withOpenAICompatibleDiscoveryTimeout(options.timeoutMs, fetchModels)
+		: fetchModels();
 }
 
 export async function fetchLiteLLMRichModels<TApi extends Api>(
@@ -6544,13 +6512,7 @@ export function nanoGptModelManagerOptions(
 						const maxOutputTokens = entry.max_output_tokens ?? entry.max_completion_tokens;
 						const mapped = mapWithBundledReference(
 							{ ...entry, max_completion_tokens: maxOutputTokens },
-							reference
-								? defaults
-								: {
-										...defaults,
-										contextWindow: toPositiveNumberOrNull(entry.context_length),
-										maxTokens: toPositiveNumberOrNull(maxOutputTokens),
-									},
+							defaults,
 							reference,
 						);
 						return {
@@ -6783,7 +6745,7 @@ export function githubCopilotModelManagerOptions(config?: GithubCopilotModelMana
 			fetchDynamicModels: async () => {
 				const fetchImpl = discoveryFetch(config?.fetch);
 				const requestBaseUrl = isPersonalGitHubCopilotBaseUrl(baseUrl)
-					? ((await withCatalogDiscoveryTimeout(DEFAULT_OPENAI_COMPATIBLE_DISCOVERY_TIMEOUT_MS, signal =>
+					? ((await withOpenAICompatibleDiscoveryTimeout(DEFAULT_OPENAI_COMPATIBLE_DISCOVERY_TIMEOUT_MS, signal =>
 							discoverGitHubCopilotApiEndpoint(apiKey, fetchImpl, signal),
 						)) ?? baseUrl)
 					: baseUrl;
@@ -7794,6 +7756,8 @@ const MODELS_DEV_PROVIDER_DESCRIPTORS_SPECIALIZED: readonly ModelsDevProviderDes
 			return { api: "openai-completions" as const, baseUrl: ZENMUX_OPENAI_BASE_URL };
 		},
 	}),
+	// --- Modular ---
+	openAiCompletionsDescriptor("modular", "modular", "https://api.modular.com/v1"),
 ];
 /** All provider descriptors for models.dev data mapping in generate-models.ts. */
 export const MODELS_DEV_PROVIDER_DESCRIPTORS: readonly ModelsDevProviderDescriptor[] = [
@@ -8238,3 +8202,40 @@ export function singularityApiTechModelManagerOptions(
 ): ModelManagerOptions<Api> {
 	return singularityApiModelManagerOptions("singularityapi-tech", SINGULARITYAPI_TECH_API_BASE_URL, config);
 }
+
+// ---------------------------------------------------------------------------
+// Scaleway
+// ---------------------------------------------------------------------------
+
+/**
+ * Scaleway discovery configuration: the API key plus optional base-URL and
+ * fetch overrides. Consumed by {@link scalewayModelManagerOptions}, and
+ * exported for extensions and tests that construct the manager directly.
+ */
+export interface ScalewayModelManagerConfig {
+	apiKey?: string;
+	baseUrl?: string;
+	fetch?: FetchImpl;
+}
+
+/**
+ * Scaleway model manager: plain OpenAI-compatible chat completions at
+ * `api.scaleway.ai/v1` (fr-par serverless Generative APIs). `/v1/models`
+ * returns bare ids, so the authored seed in `providers/scaleway.kdl` is the
+ * reference each live row merges names, modalities, prices, and limits from;
+ * a successful fetch is the complete roster and prunes retired rows.
+ */
+export function scalewayModelManagerOptions(
+	config?: ScalewayModelManagerConfig,
+): ModelManagerOptions<"openai-completions"> {
+	return createOpenAICompatibleModelManagerOptions({
+		api: "openai-completions",
+		providerId: "scaleway",
+		defaultBaseUrl: "https://api.scaleway.ai/v1",
+		config,
+		dynamicModelsAuthoritative: true,
+		requireApiKey: true,
+		filterModel: (_entry, model) => !isExcludedModel("scaleway", model.id),
+		mapModel: mapWithBundledReference,
+	});
+}

package/src/types.ts

@@ -1,5 +1,6 @@
 import type { ModelIdentity } from "./compat/types";
 import type { Effort } from "./effort";
+import type { DatedCostCard } from "./pricing";
 
 // Re-exported from @oh-my-pi/pi-utils so the whole workspace shares one
 // `fetch`-compatible signature (tls-fetch's wrappers produce/accept it).
@@ -1477,8 +1478,8 @@ export interface ModelCost extends TokenCost {
 /**
  * Exact local content tokenizer family for a model.
  *
- * Absent means no first-party local tokenizer is known and consumers retain
- * their estimate/default-tokenizer policy. The values name tokenizer
+ * Absent means no first-party local tokenizer is known and consumers count
+ * in o200k_base. The values name tokenizer
  * generations rather than providers: DeepSeek V3 through V4 share
  * `"deepseek-v3"`; Kimi K2 through K3 share `"kimi-k2"`.
  */
@@ -1491,6 +1492,7 @@ export type ModelTokenizer =
 	| "deepseek-v3"
 	| "kimi-k2"
 	| "glm5"
+	| "cl100k"
 	| "o200k";
 
 /** One account's discovered entitlements on a model; see {@link Model.accountAccess}. */
@@ -1686,8 +1688,12 @@ export interface Model<TApi extends Api = Api> {
 	factoryDroidApiProviders?: string[];
 	/** Factory Droid: `configured_order` when live routing chose the rotation; absent means registry order. */
 	factoryDroidRoutingSource?: "configured_order";
+	/** Factory Droid: server-side refusal fallbacks the account can run; absent when none are named. */
+	factoryDroidRefusalFallbacks?: string[];
 	/** Factory Droid Standard Credits base rate (relative per-token weight, not dollars). */
 	factoryDroidCredits?: number;
+	/** Factory Droid dated `credit-rates` cards, ascending; `getFactoryDroidCreditRate` picks the one due now. */
+	factoryDroidCreditCards?: readonly DatedCostCard[];
 	/** Canonical organization resolved during Factory discovery. */
 	factoryDroidOrgId?: string;
 	cost: ModelCost;
@@ -1760,8 +1766,9 @@ export interface Model<TApi extends Api = Api> {
 	useResponsesLite?: boolean;
 	/**
 	 * Service-tier ids the provider advertises for this model (Codex discovery
-	 * `service_tiers[].id`, e.g. `priority`, `ultrafast`). Absent when the
-	 * provider publishes no per-model tier list.
+	 * `service_tiers[].id`, or an OpenAI-compatible `/v1/models` row's
+	 * `service_tiers`, e.g. `priority`, `ultrafast`). Absent when the provider
+	 * publishes no per-model tier list.
 	 */
 	serviceTiers?: readonly string[];
 	/** Preferred model to switch to when context promotion is triggered (model id or provider/id). */

package/src/utils.ts

@@ -40,6 +40,24 @@ export function toBoolean(value: unknown): boolean | undefined {
 	return typeof value === "boolean" ? value : undefined;
 }
 
+/**
+ * Service-tier ids from a discovery row's `service_tiers` array, deduplicated
+ * in advertisement order. Entries may be codex-rs `{ id, name, description }`
+ * objects or bare id strings (OpenAI-compatible relays). Returns `undefined`
+ * when the value is not an array; an empty array is kept, since it is a
+ * statement that the model offers no optional tier.
+ */
+export function parseServiceTierIds(value: unknown): string[] | undefined {
+	if (!Array.isArray(value)) return undefined;
+	const ids: string[] = [];
+	for (const tier of value) {
+		const raw = tier !== null && typeof tier === "object" && "id" in tier ? tier.id : tier;
+		const id = typeof raw === "string" ? raw.trim() : "";
+		if (id && !ids.includes(id)) ids.push(id);
+	}
+	return ids;
+}
+
 export function isAnthropicOAuthToken(key: string): boolean {
 	return key.includes("sk-ant-oat");
 }

package/src/wire/factory-droid.ts

@@ -124,7 +124,8 @@ export function factoryDroidModelCacheProviderId(options: { apiKey?: string } &
 	}
 	const region = options.region === "eu" ? "eu" : "global";
 	const scope = `${credentialScope}\u0000${options.orgId ?? ""}\u0000${region}\u0000${options.inferenceRegion ?? region}`;
-	// v4: rows cached before credits became a scalar base rate carry the old
-	// `{ input, output, cacheRead }` object and must be refetched.
-	return `factory-droid:models-v4:${Bun.hash(scope).toString(36)}`;
+	// v5: rows cached before discovery recorded the account's refusal fallbacks
+	// would let the request path send every fallback the rules name, so they
+	// must be refetched.
+	return `factory-droid:models-v5:${Bun.hash(scope).toString(36)}`;
 }