@oh-my-pi/pi-catalog 17.2.15 → 17.3.0

package/CHANGELOG.md

@@ -2,6 +2,29 @@
 
 ## [Unreleased]
 
+## [17.3.0] - 2026-08-13
+
+### Breaking Changes
+
+- Removed `OpenAICompat.enableGeminiThinkingLoopGuard`; thinking-loop eligibility is derived solely from the `model.id` family.
+
+### Added
+
+- Added first-party OpenAI Daybreak Blue, Daybreak Red, and GPT-5.6 Cyber models with full support for their documented API pricing (including long-context rates above 272K input), token limits, tools, and reasoning effort controls (off/low/medium/high/xhigh/max).
+- Added calculateUncachedInputCost() to calculate prompt pricing against active context-length tiers without prompt caching.
+
+### Fixed
+
+- Fixed Anthropic cache-write pricing to correctly honor mixed 5-minute and 1-hour TTL usage instead of incorrectly charging all writes at the 5-minute rate.
+- Fixed Ollama Cloud DeepSeek V4 Flash and older reasoners to correctly apply the DeepSeek effort contract (e.g., low/high/max) instead of the generic effort ladder.
+- Added a default request timeout to OpenAI-compatible model discovery to prevent stalled provider endpoints from hanging startup indefinitely.
+- Fixed Anthropic cache-write pricing to honor mixed 5-minute and 1-hour TTL usage instead of charging every write at the 5-minute rate.
+- Fixed Ollama Cloud DeepSeek V4 Flash (including dated/preview tags like `deepseek-v4-flash:0731`) exposing the generic `minimal`/`low`/`medium`/`high`/`xhigh` effort ladder without `max`; the `ollama-chat` transport now applies the DeepSeek effort contract (Flash → `low`/`high`/`max`, older reasoners → `high`/`max`), matching the direct API and every other host ([#8334](https://github.com/can1357/oh-my-pi/issues/8334)).
+- Exposed the `low` reasoning-effort tier for DeepSeek V4 Pro on the direct API and faithful aggregator routes, matching DeepSeek's updated API contract advertising `reasoning_effort` `low`/`high`/`max` for both V4 SKUs; OpenRouter's non-Flash route still exposes only `high`, and the older V3.x/R1 reasoners remain `high`/`max` ([#8405](https://github.com/can1357/oh-my-pi/issues/8405)).
+- Bounded OpenAI-compatible model discovery with a default request timeout so a stalled provider `/models` endpoint can no longer hang startup indefinitely in `resolveModelDiscoveryFallback` ([#8315](https://github.com/can1357/oh-my-pi/issues/8315)).
+- Fixed Codex-discovered `gpt-daybreak-*` aliases being treated as unknown models, restoring the GPT-5.6 `low`/`medium`/`high`/`xhigh`/`max` effort ladder and its 372K fallback only when the Codex registry omits `context_window`.
+- Fixed first-party OpenAI GPT-5.6 aliases to preserve wire-level `off` through generated pro aliases and to price requests above 272K input at each SKU's documented long-context rates.
+
 ## [17.2.15] - 2026-08-12
 
 ### Fixed

package/dist/types/discovery/openai-compatible.d.ts

@@ -1,4 +1,15 @@
 import type { Api, FetchImpl, ModelSpec, Provider } from "../types.js";
+/**
+ * Default hard deadline applied to an OpenAI-compatible `/models` probe when
+ * the caller supplies neither an `AbortSignal` nor an explicit `timeoutMs`.
+ *
+ * Built-in provider model managers (openrouter, xAI, DeepSeek, …) call
+ * {@link fetchOpenAICompatibleModels} with no timeout, so without this bound a
+ * stalled endpoint left the request pending forever and blocked startup's
+ * awaited `resolveModelDiscoveryFallback` discovery pass indefinitely
+ * (issue #8315). 10s matches the coding-agent's remote-discovery budget.
+ */
+export declare const DEFAULT_OPENAI_COMPATIBLE_DISCOVERY_TIMEOUT_MS = 10000;
 /**
  * Minimal OpenAI-style model entry shape consumed by discovery.
  *
@@ -50,7 +61,11 @@ export interface FetchOpenAICompatibleModelsOptions<TApi extends Api> {
     headers?: Record<string, string>;
     /** Optional AbortSignal for request cancellation; caller owns its lifecycle. */
     signal?: AbortSignal;
-    /** Optional cancellable request timeout used when `signal` is omitted. */
+    /**
+     * Optional cancellable request timeout used when `signal` is omitted.
+     * Defaults to {@link DEFAULT_OPENAI_COMPATIBLE_DISCOVERY_TIMEOUT_MS} so a
+     * stalled endpoint can never hang discovery indefinitely.
+     */
     timeoutMs?: number;
     /** Optional fetch implementation override for testing/custom runtimes. */
     fetch?: FetchImpl;

package/dist/types/identity/family.d.ts

@@ -36,13 +36,19 @@ export declare const isGemmaModelId: (modelId: string) => boolean;
 export declare const isDeepseekModelIdOrName: (modelId: string) => boolean;
 /**
  * DeepSeek V4 Flash SKU in any host/namespace form (`deepseek-v4-flash`, dated
- * `deepseek-v4-flash-0731`, `deepseek-ai/DeepSeek-V4-Flash`). Flash is the only
- * V4 model whose `reasoning_effort` accepts the `low` tier; V4 Pro tops out at
- * `high`/`max`. See https://api-docs.deepseek.com/api/create-chat-completion.
+ * `deepseek-v4-flash-0731`, `deepseek-ai/DeepSeek-V4-Flash`). Both V4 SKUs
+ * (Flash and Pro) accept the `low` reasoning_effort tier; this predicate keeps
+ * Flash distinguishable from Pro where a host quirk splits them (e.g.
+ * OpenRouter exposes `low` on Flash but only `high` on non-Flash V4).
+ * See https://api-docs.deepseek.com/api/create-chat-completion.
  */
 export declare const isDeepseekV4FlashModelId: (modelId: string) => boolean;
 /** Xiaomi MiMo family by id or display name. */
 export declare const isMimoModelIdOrName: (modelId: string) => boolean;
+/** Gemini family ids in any namespace form (`gemini-*`, `google/gemini-*`, `openrouter/google/gemini-…`). */
+export declare const isGeminiModelId: (modelId: string) => boolean;
+/** Grok family ids across namespace and delimiter forms (`grok-*`, `cursor-grok-*`, `xai/grok-*`). */
+export declare const isGrokModelId: (modelId: string) => boolean;
 /**
  * Grok SKUs that expose the wire `reasoning.effort` dial. Other Grok reasoners
  * (e.g. `grok-build`, `grok-4.20-0309-reasoning`) think natively but reject the

package/dist/types/models.d.ts

@@ -4,6 +4,8 @@ export type GeneratedProvider = keyof typeof MODELS;
 export declare function getBundledModel<TApi extends Api = Api>(provider: GeneratedProvider, modelId: string): Model<TApi>;
 export declare function getBundledProviders(): KnownProvider[];
 export declare function getBundledModels(provider: GeneratedProvider): Model<Api>[];
+/** Price a prompt as fully uncached input under its active context-length tier. */
+export declare function calculateUncachedInputCost(cost: Model["cost"], promptInputTokens: number): number;
 export declare function calculateCost<TApi extends Api>(model: Model<TApi>, usage: Usage): Usage["cost"];
 /**
  * Check if two models are equal by comparing both their id and provider.

package/dist/types/provider-models/openai-compat.d.ts

@@ -61,12 +61,41 @@ export interface UmansModelManagerConfig {
     fetch?: FetchImpl;
 }
 export declare function umansModelManagerOptions(config?: UmansModelManagerConfig): ModelManagerOptions<"anthropic-messages">;
+export declare const OPENAI_GPT_56_LONG_CONTEXT_COSTS: {
+    readonly luna: {
+        readonly inputThreshold: 272000;
+        readonly input: 0.4;
+        readonly output: 1.8;
+        readonly cacheRead: 0.04;
+        readonly cacheWrite: 0.5;
+    };
+    readonly sol: {
+        readonly inputThreshold: 272000;
+        readonly input: 10;
+        readonly output: 45;
+        readonly cacheRead: 1;
+        readonly cacheWrite: 12.5;
+    };
+    readonly terra: {
+        readonly inputThreshold: 272000;
+        readonly input: 4;
+        readonly output: 18;
+        readonly cacheRead: 0.4;
+        readonly cacheWrite: 5;
+    };
+};
 export interface OpenAIModelManagerConfig {
     apiKey?: string;
     baseUrl?: string;
     fetch?: FetchImpl;
 }
 export declare function openaiModelManagerOptions(config?: OpenAIModelManagerConfig): ModelManagerOptions<"openai-responses">;
+/**
+ * Daybreak models are approval-gated first-party Responses models that are not
+ * yet present in stencil.so. Seed the documented aliases and current Cyber
+ * snapshot so fresh installs expose them without credentialed discovery.
+ */
+export declare const OPENAI_DAYBREAK_CURATED_FALLBACK_MODELS: readonly ModelSpec<"openai-responses">[];
 /**
  * Re-derive the generated pro-reasoning aliases (`gpt-5.6-*-pro`) for the
  * first-party `openai` gpt-5.6 rows. Each alias inherits the base row's

package/dist/types/types.d.ts

@@ -122,7 +122,7 @@ export interface Usage {
     };
 }
 export type OpenAIReasoningFormat = "openai" | "openrouter" | "zai" | "kimi" | "qwen" | "qwen-chat-template";
-export type OpenAIReasoningDisableMode = "omit" | "lowest-effort" | "openrouter-enabled-false" | "zai-thinking-disabled" | "qwen-enable-thinking-false" | "qwen-template-false";
+export type OpenAIReasoningDisableMode = "omit" | "lowest-effort" | "none-effort" | "openrouter-enabled-false" | "zai-thinking-disabled" | "qwen-enable-thinking-false" | "qwen-template-false";
 export type OpenAIStreamMarkupHealingPattern = "kimi" | "dsml" | "thinking";
 /**
  * Compatibility settings for openai-completions API.
@@ -152,13 +152,6 @@ export interface OpenAICompat {
     reasoningEffortMap?: Partial<Record<Effort, string>>;
     /** Whether the provider supports `stream_options: { include_usage: true }` for token usage in streaming responses. Default: true. */
     supportsUsageInStreaming?: boolean;
-    /**
-     * Enable the Gemini thinking-loop guard (pi-ai stream layer) for this model.
-     * Defaults to true when the model id classifies as the gemini family. Set
-     * explicitly to cover an opaque OpenAI-compat proxy alias (e.g. `my-model`)
-     * that routes to Gemini, or to false to opt a gemini-family id out.
-     */
-    enableGeminiThinkingLoopGuard?: boolean;
     /** Which field to use for max tokens. Default: auto-detected from URL. */
     maxTokensField?: "max_completion_tokens" | "max_tokens";
     /** Whether tool results require the `name` field. Default: auto-detected from URL. */
@@ -582,8 +575,6 @@ export interface ResolvedOpenAISharedCompat {
     isOpenRouterHost: boolean;
     /** Whether this endpoint needs a max-token field even when caller did not set one. */
     alwaysSendMaxTokens: boolean;
-    /** See {@link OpenAICompat.enableGeminiThinkingLoopGuard}. Set by the builder from the family classifier. */
-    enableGeminiThinkingLoopGuard?: boolean;
     openRouterRouting?: OpenAICompat["openRouterRouting"];
     /** Provider-specific wire model-id transform applied to the base id. */
     wireModelIdMode: "raw" | "firepass" | "fireworks" | "openrouter";
@@ -596,7 +587,7 @@ export interface ResolvedOpenAISharedCompat {
  * `buildModel`; request handlers read fields and never detect, resolve, or
  * allocate.
  */
-export type ResolvedOpenAICompat = ResolvedOpenAISharedCompat & Required<Omit<OpenAICompat, "supportsDeveloperRole" | "supportsReasoningEffort" | "reasoningEffortMap" | "supportsReasoningParams" | "supportsSamplingParams" | "thinkingFormat" | "kimiApiFormat" | "reasoningDisableMode" | "omitReasoningEffort" | "includeEncryptedReasoning" | "filterReasoningHistory" | "disableReasoningOnForcedToolChoice" | "disableReasoningOnToolChoice" | "supportsToolChoice" | "supportsForcedToolChoice" | "supportsNamedToolChoice" | "reasoningContentField" | "requiresReasoningContentForToolCalls" | "requiresReasoningContentForAllAssistantTurns" | "allowsSyntheticReasoningContentForToolCalls" | "replayReasoningContent" | "qwenPreserveThinking" | "requiresThinkingAsText" | "requiresMistralToolIds" | "requiresToolResultName" | "requiresAssistantAfterToolResult" | "requiresAssistantContentForToolCalls" | "stripDeepseekSpecialTokens" | "streamMarkupHealingPattern" | "reasoningDeltasMayBeCumulative" | "emptyLengthFinishIsContextError" | "usesOpenAIToolCallIdLimit" | "promptCacheSessionHeader" | "supportsPromptCacheBreakpoints" | "promptCacheBreakpointTtl" | "openRouterRouting" | "isOpenRouterHost" | "supportsStrictMode" | "supportsLongPromptCacheRetention" | "alwaysSendMaxTokens" | "wireModelIdMode" | "vercelGatewayRouting" | "extraBody" | "toolStrictMode" | "toolSchemaFlavor" | "streamFirstEventTimeoutMs" | "streamIdleTimeoutMs" | "cacheControlFormat" | "thinkingKeep" | "strictResponsesPairing" | "supportsImageDetailOriginal" | "enableGeminiThinkingLoopGuard" | "whenThinking">> & {
+export type ResolvedOpenAICompat = ResolvedOpenAISharedCompat & Required<Omit<OpenAICompat, "supportsDeveloperRole" | "supportsReasoningEffort" | "reasoningEffortMap" | "supportsReasoningParams" | "supportsSamplingParams" | "thinkingFormat" | "kimiApiFormat" | "reasoningDisableMode" | "omitReasoningEffort" | "includeEncryptedReasoning" | "filterReasoningHistory" | "disableReasoningOnForcedToolChoice" | "disableReasoningOnToolChoice" | "supportsToolChoice" | "supportsForcedToolChoice" | "supportsNamedToolChoice" | "reasoningContentField" | "requiresReasoningContentForToolCalls" | "requiresReasoningContentForAllAssistantTurns" | "allowsSyntheticReasoningContentForToolCalls" | "replayReasoningContent" | "qwenPreserveThinking" | "requiresThinkingAsText" | "requiresMistralToolIds" | "requiresToolResultName" | "requiresAssistantAfterToolResult" | "requiresAssistantContentForToolCalls" | "stripDeepseekSpecialTokens" | "streamMarkupHealingPattern" | "reasoningDeltasMayBeCumulative" | "emptyLengthFinishIsContextError" | "usesOpenAIToolCallIdLimit" | "promptCacheSessionHeader" | "supportsPromptCacheBreakpoints" | "promptCacheBreakpointTtl" | "openRouterRouting" | "isOpenRouterHost" | "supportsStrictMode" | "supportsLongPromptCacheRetention" | "alwaysSendMaxTokens" | "wireModelIdMode" | "vercelGatewayRouting" | "extraBody" | "toolStrictMode" | "toolSchemaFlavor" | "streamFirstEventTimeoutMs" | "streamIdleTimeoutMs" | "cacheControlFormat" | "thinkingKeep" | "strictResponsesPairing" | "supportsImageDetailOriginal" | "whenThinking">> & {
     vercelGatewayRouting?: OpenAICompat["vercelGatewayRouting"];
     extraBody?: OpenAICompat["extraBody"];
     cacheControlFormat?: OpenAICompat["cacheControlFormat"];
@@ -681,6 +672,25 @@ export interface RemoteCompactionConfig<TApi extends Api = Api> {
     /** Model id sent to the compaction endpoint when it differs from the active model id. */
     model?: string;
 }
+/** Per-million-token rates for one model pricing tier. */
+export interface TokenCost {
+    input: number;
+    output: number;
+    cacheRead: number;
+    cacheWrite: number;
+}
+/**
+ * Rates applied to the full request when its prompt exceeds `inputThreshold`.
+ * Prompt input is the sum of uncached, cached-read, cache-write, and
+ * provider-orchestration input tokens.
+ */
+export interface LongContextTokenCost extends TokenCost {
+    inputThreshold: number;
+}
+/** Base token rates plus an optional long-context tier. */
+export interface ModelCost extends TokenCost {
+    longContext?: LongContextTokenCost;
+}
 export interface Model<TApi extends Api = Api> {
     id: string;
     /**
@@ -725,12 +735,7 @@ export interface Model<TApi extends Api = Api> {
     gitlabDuoWorkflowRootNamespaceId?: string;
     /** Cursor `max_mode` request flag returned by `GetUsableModels` for premium models that require max mode. */
     cursorMaxMode?: boolean;
-    cost: {
-        input: number;
-        output: number;
-        cacheRead: number;
-        cacheWrite: number;
-    };
+    cost: ModelCost;
     /** Premium Copilot requests charged per user-initiated request (defaults to 1). */
     premiumMultiplier?: number;
     contextWindow: number | null;

package/package.json

@@ -1,7 +1,7 @@
 {
 	"type": "module",
 	"name": "@oh-my-pi/pi-catalog",
-	"version": "17.2.15",
+	"version": "17.3.0",
 	"description": "Model catalog for omp: bundled model database, provider discovery descriptors, model identity, classification, and equivalence",
 	"homepage": "https://omp.sh",
 	"author": "Can Boluk",
@@ -35,11 +35,11 @@
 	},
 	"dependencies": {
 		"@bufbuild/protobuf": "^2.12.1",
-		"@oh-my-pi/omptype": "17.2.15",
-		"@oh-my-pi/pi-utils": "17.2.15"
+		"@oh-my-pi/omptype": "17.3.0",
+		"@oh-my-pi/pi-utils": "17.3.0"
 	},
 	"devDependencies": {
-		"@oh-my-pi/pi-ai": "17.2.15",
+		"@oh-my-pi/pi-ai": "17.3.0",
 		"@types/bun": "^1.3.14"
 	},
 	"engines": {

package/src/build.ts

@@ -14,12 +14,11 @@ import { buildAnthropicCompat } from "./compat/anthropic";
 import { buildBedrockCompat } from "./compat/bedrock";
 import { buildDevinCompat } from "./compat/devin";
 import { buildOpenAICompat, buildOpenAIResponsesCompat, buildOpenRouterCompat } from "./compat/openai";
+import { bareModelId, parseOpenAIModel, semverGte } from "./identity/classify";
 import { resolveModelThinking } from "./model-thinking";
 import type { Api, CompatOf, Model, ModelSpec } from "./types";
 import { cleanModelName } from "./utils";
 
-const OPENAI_GA_COMPUTER_MODEL_RE = /^gpt-5\.(?:[4-9]|[1-9]\d)(?:[.-]|$)/i;
-
 function isDirectOpenAIResponsesEndpoint(spec: ModelSpec<Api>): boolean {
 	if (spec.api === "openai-responses") {
 		if (spec.provider !== "openai") return false;
@@ -55,7 +54,8 @@ function explicitComputerUseConfig(spec: ModelSpec<Api>): boolean | undefined {
 function supportsOpenAIGAComputerUse(spec: ModelSpec<Api>, explicitSupport: boolean | undefined): boolean {
 	if (explicitSupport !== undefined) return explicitSupport;
 	if (!isDirectOpenAIResponsesEndpoint(spec)) return false;
-	return OPENAI_GA_COMPUTER_MODEL_RE.test(spec.requestModelId ?? spec.id);
+	const parsed = parseOpenAIModel(bareModelId(spec.requestModelId ?? spec.id));
+	return parsed !== null && semverGte(parsed.version, "5.4");
 }
 
 export function buildModel<TApi extends Api>(spec: ModelSpec<TApi>): Model<TApi> {

package/src/compat/openai.ts

@@ -22,7 +22,6 @@ import {
 	isMimoModelIdOrName,
 	isOpenAISamplingRestrictedModelId,
 	isQwenModelId,
-	modelFamilyToken,
 } from "../identity/family";
 import type {
 	ModelSpec,
@@ -474,10 +473,6 @@ export function buildOpenAICompat(spec: ModelSpec<"openai-completions">): Resolv
 		supportsSamplingParams: !isOpenAISamplingRestrictedModelId(spec.id),
 		reasoningEffortMap: {},
 		supportsUsageInStreaming: !isCerebras,
-		// pi-ai's thinking-loop guard is gemini-only; default the flag from the
-		// family classifier so OpenAI-compat proxies serving Gemini are covered.
-		// An opaque alias can opt in via `compat.enableGeminiThinkingLoopGuard`.
-		enableGeminiThinkingLoopGuard: modelFamilyToken(spec.id) === "gemini",
 		// Kimi (including via OpenRouter and Fireworks router-form IDs such as
 		// `accounts/fireworks/routers/kimi-*`) calculates TPM rate limits based on
 		// max_tokens, not actual output. The official Kimi K2 model guidance
@@ -749,7 +744,6 @@ export function buildOpenAIResponsesCompat(spec: OpenAIResponsesSpecLike): Resol
 		// lands on Moonshot's MFJS validator.
 		toolSchemaFlavor: isKimiModel ? "moonshot-mfjs" : undefined,
 		alwaysSendMaxTokens: spec.id ? isKimiModelId(spec.id) : false,
-		enableGeminiThinkingLoopGuard: modelFamilyToken(spec.id ?? "") === "gemini",
 		supportsObfuscationOptOut: isOpenAIUrl || spec.provider === "openai",
 		stripDeepseekSpecialTokens:
 			Boolean(id) && isDeepseekModelIdOrName(id) && (spec.provider === "nvidia" || spec.provider === "deepseek"),

package/src/discovery/openai-compatible.ts

@@ -4,6 +4,18 @@ import { discoveryFetch } from "../utils";
 
 const MODELS_PATH = "/models";
 
+/**
+ * Default hard deadline applied to an OpenAI-compatible `/models` probe when
+ * the caller supplies neither an `AbortSignal` nor an explicit `timeoutMs`.
+ *
+ * Built-in provider model managers (openrouter, xAI, DeepSeek, …) call
+ * {@link fetchOpenAICompatibleModels} with no timeout, so without this bound a
+ * stalled endpoint left the request pending forever and blocked startup's
+ * awaited `resolveModelDiscoveryFallback` discovery pass indefinitely
+ * (issue #8315). 10s matches the coding-agent's remote-discovery budget.
+ */
+export const DEFAULT_OPENAI_COMPATIBLE_DISCOVERY_TIMEOUT_MS = 10_000;
+
 /**
  * Uses a cancellable timer rather than the native abort-timeout helper so
  * successful fast discovery requests do not leave armed timeout signals for
@@ -96,7 +108,11 @@ export interface FetchOpenAICompatibleModelsOptions<TApi extends Api> {
 	headers?: Record<string, string>;
 	/** Optional AbortSignal for request cancellation; caller owns its lifecycle. */
 	signal?: AbortSignal;
-	/** Optional cancellable request timeout used when `signal` is omitted. */
+	/**
+	 * Optional cancellable request timeout used when `signal` is omitted.
+	 * Defaults to {@link DEFAULT_OPENAI_COMPATIBLE_DISCOVERY_TIMEOUT_MS} so a
+	 * stalled endpoint can never hang discovery indefinitely.
+	 */
 	timeoutMs?: number;
 	/** Optional fetch implementation override for testing/custom runtimes. */
 	fetch?: FetchImpl;
@@ -164,9 +180,10 @@ export async function fetchOpenAICompatibleModels<TApi extends Api>(
 	const payload =
 		options.signal !== undefined
 			? await fetchPayload(options.signal)
-			: options.timeoutMs !== undefined
-				? await withOpenAICompatibleDiscoveryTimeout(options.timeoutMs, fetchPayload)
-				: await fetchPayload();
+			: await withOpenAICompatibleDiscoveryTimeout(
+					options.timeoutMs ?? DEFAULT_OPENAI_COMPATIBLE_DISCOVERY_TIMEOUT_MS,
+					fetchPayload,
+				);
 	if (payload === null) {
 		return null;
 	}

package/src/identity/classify.ts

@@ -122,16 +122,31 @@ export const parseAnthropicModel = parser((modelId): AnthropicModel | null => {
 	return { family: "anthropic", kind: kind as AnthropicKind, version };
 });
 
+/**
+ * Rolling OpenAI aliases inherit wire capabilities from their current default
+ * snapshots. Keep this map aligned with the model docs when an alias advances.
+ */
+const OPENAI_ALIAS_VERSIONS: Readonly<Record<string, string>> = {
+	"daybreak-blue-latest": "5.6",
+	"gpt-daybreak-blue-latest": "5.6",
+	"daybreak-red-latest": "5.6",
+	"gpt-daybreak-red-latest": "5.6",
+};
+
 export const parseOpenAIModel = parser((modelId): OpenAIModel | null => {
-	const match = /gpt-(\d+(?:\.\d+){0,2})(?:-(codex-spark|codex-mini|codex-max|codex|mini|max|nano))?\b/.exec(modelId);
-	if (!match) {
+	const aliasVersion = OPENAI_ALIAS_VERSIONS[modelId];
+	const match = aliasVersion
+		? null
+		: /gpt-(\d+(?:\.\d+){0,2})(?:-(codex-spark|codex-mini|codex-max|codex|mini|max|nano))?\b/.exec(modelId);
+	const versionInput = aliasVersion ?? match?.[1];
+	if (!versionInput) {
 		return null;
 	}
-	const version = parseSemVer(match[1]);
+	const version = parseSemVer(versionInput);
 	if (!version) {
 		return null;
 	}
-	return { family: "openai", variant: (match[2] as OpenAIVariant | undefined) ?? "base", version };
+	return { family: "openai", variant: (match?.[2] as OpenAIVariant | undefined) ?? "base", version };
 });
 
 /**

package/src/identity/family.ts

@@ -85,9 +85,11 @@ export const isDeepseekModelIdOrName = memo((value: string): boolean => {
 
 /**
  * DeepSeek V4 Flash SKU in any host/namespace form (`deepseek-v4-flash`, dated
- * `deepseek-v4-flash-0731`, `deepseek-ai/DeepSeek-V4-Flash`). Flash is the only
- * V4 model whose `reasoning_effort` accepts the `low` tier; V4 Pro tops out at
- * `high`/`max`. See https://api-docs.deepseek.com/api/create-chat-completion.
+ * `deepseek-v4-flash-0731`, `deepseek-ai/DeepSeek-V4-Flash`). Both V4 SKUs
+ * (Flash and Pro) accept the `low` reasoning_effort tier; this predicate keeps
+ * Flash distinguishable from Pro where a host quirk splits them (e.g.
+ * OpenRouter exposes `low` on Flash but only `high` on non-Flash V4).
+ * See https://api-docs.deepseek.com/api/create-chat-completion.
  */
 export const isDeepseekV4FlashModelId = memo((modelId: string): boolean => {
 	return bareModelId(modelId).toLowerCase().includes("deepseek-v4-flash");
@@ -98,6 +100,16 @@ export const isMimoModelIdOrName = memo((value: string): boolean => {
 	return value.toLowerCase().includes("mimo");
 });
 
+/** Gemini family ids in any namespace form (`gemini-*`, `google/gemini-*`, `openrouter/google/gemini-…`). */
+export const isGeminiModelId = memo((modelId: string): boolean => {
+	return /(^|\/)gemini[-.]?/i.test(modelId);
+});
+
+/** Grok family ids across namespace and delimiter forms (`grok-*`, `cursor-grok-*`, `xai/grok-*`). */
+export const isGrokModelId = memo((modelId: string): boolean => {
+	return /(?:^|[./_-])grok(?:[-.]|$)/i.test(modelId);
+});
+
 const GROK_EFFORT_CAPABLE_PREFIXES = ["grok-3-mini", "grok-4.20-multi-agent", "grok-4.3", "grok-4.5"] as const;
 
 /**
@@ -259,12 +271,14 @@ export const modelFamilyToken = memo((modelId: string): string => {
 	const parsed = parseKnownModel(modelId);
 	if (parsed.family !== "unknown") return parsed.family;
 	if (isClaudeModelId(modelId) || isAnthropicNamespacedModelId(modelId)) return "anthropic";
+	if (isGeminiModelId(modelId)) return "gemini";
+	if (isGrokModelId(modelId)) return "grok";
+	if (isDeepseekModelIdOrName(modelId)) return "deepseek";
 	if (isOpenAIModelId(modelId)) return "openai";
 	if (isKimiModelId(modelId)) return "kimi";
 	if (isQwenModelId(modelId)) return "qwen";
 	if (isMinimaxM2FamilyModelId(modelId) || isMinimaxM3FamilyModelId(modelId)) return "minimax";
 	if (isOpenAIGptOssModelId(modelId)) return "gpt-oss";
-	if (isDeepseekModelIdOrName(modelId)) return "deepseek";
 	if (isMimoModelIdOrName(modelId)) return "mimo";
 	if (isGemmaModelId(modelId)) return "gemma";
 	if (parseGlmModel(bareModelId(modelId))) return "glm";

package/src/model-manager.ts

@@ -1,7 +1,7 @@
 import { buildModel } from "./build";
 import { readModelCache, writeModelCache } from "./model-cache";
 import { type GeneratedProvider, getBundledModels } from "./models";
-import type { Api, Model, ModelSpec, Provider } from "./types";
+import type { Api, Model, ModelCost, ModelSpec, Provider, TokenCost } from "./types";
 import { isRecord } from "./utils";
 import { collapseBuiltModelVariants } from "./variant-collapse";
 
@@ -510,6 +510,7 @@ function mergeDynamicModel<TApi extends Api>(existingModel: Model<TApi>, dynamic
 	const reasoning = dynamicReasoningAuthoritative
 		? dynamicModel.reasoning
 		: existingModel.reasoning || dynamicModel.reasoning;
+	const longContextCost = dynamicModel.cost.longContext ?? existingModel.cost.longContext;
 	// Re-build from spec stage: sparse compat comes from `compatConfig` (the
 	// verbatim override vocabulary), never the resolved `compat` record.
 	return buildModel({
@@ -523,6 +524,7 @@ function mergeDynamicModel<TApi extends Api>(existingModel: Model<TApi>, dynamic
 			output: preferDiscoveryCost(dynamicModel.cost.output, existingModel.cost.output),
 			cacheRead: preferDiscoveryCost(dynamicModel.cost.cacheRead, existingModel.cost.cacheRead),
 			cacheWrite: preferDiscoveryCost(dynamicModel.cost.cacheWrite, existingModel.cost.cacheWrite),
+			...(longContextCost ? { longContext: longContextCost } : {}),
 		},
 		contextWindow: preferDiscoveryLimit(dynamicModel.contextWindow, existingModel.contextWindow),
 		maxTokens: preferDiscoveryLimit(dynamicModel.maxTokens, existingModel.maxTokens),
@@ -640,7 +642,7 @@ function isModelInputArray(value: unknown): value is ("text" | "image")[] {
 	return true;
 }
 
-function isModelCost(value: unknown): value is Model<Api>["cost"] {
+function isTokenCost(value: unknown): value is TokenCost {
 	if (!isRecord(value)) {
 		return false;
 	}
@@ -670,3 +672,12 @@ function isModelCost(value: unknown): value is Model<Api>["cost"] {
 	}
 	return true;
 }
+
+function isModelCost(value: unknown): value is ModelCost {
+	if (!isTokenCost(value)) return false;
+	const longContext = (value as TokenCost & { longContext?: unknown }).longContext;
+	if (longContext === undefined) return true;
+	if (!isTokenCost(longContext) || !isRecord(longContext)) return false;
+	const threshold = longContext.inputThreshold;
+	return typeof threshold === "number" && threshold > 0 && threshold < Infinity;
+}

package/src/model-thinking.ts

@@ -63,9 +63,9 @@ const GEMINI_3_FLASH_EFFORTS: readonly Effort[] = [Effort.Minimal, Effort.Low, E
 const GPT_5_2_PLUS_EFFORTS: readonly Effort[] = [Effort.Low, Effort.Medium, Effort.High, Effort.XHigh];
 const GPT_5_1_CODEX_MINI_EFFORTS: readonly Effort[] = [Effort.Medium, Effort.High];
 const LOW_MEDIUM_HIGH_REASONING_EFFORTS: readonly Effort[] = [Effort.Low, Effort.Medium, Effort.High];
-/** Wire-exact `low`/`high`/`max` scale used by Kimi K3 and DeepSeek V4 Flash (direct API and aggregators). */
+/** Wire-exact `low`/`high`/`max` scale used by Kimi K3 and DeepSeek V4 (Flash and Pro, direct API and aggregators). */
 const LOW_HIGH_MAX_REASONING_EFFORTS: readonly Effort[] = [Effort.Low, Effort.High, Effort.Max];
-/** Wire-exact two-tier scale (`high`/`max`): GLM-5.2 on Z.ai/Umans/Ollama Cloud/Baseten, Sakana Fugu, DeepSeek V4 Pro. */
+/** Wire-exact two-tier scale (`high`/`max`): GLM-5.2 on Z.ai/Umans/Ollama Cloud/Baseten, Sakana Fugu, older DeepSeek reasoners (V3.x/R1). */
 const HIGH_MAX_REASONING_EFFORTS: readonly Effort[] = [Effort.High, Effort.Max];
 /** OpenRouter's DeepSeek route accepts only `high`. */
 const HIGH_ONLY_REASONING_EFFORTS: readonly Effort[] = [Effort.High];
@@ -366,14 +366,22 @@ function getModelDefinedEfforts<TApi extends Api>(
 	if (spec.provider === "ollama") {
 		return OLLAMA_REASONING_EFFORTS;
 	}
-	if (isOpenAICompatReasoningApi(spec.api) && isDeepseekReasoningModel(spec)) {
-		// DeepSeek V4 Flash accepts the wire-exact low/high/max ladder on every
-		// host — the direct API and aggregators alike (medium/xhigh map to
-		// high). V4 Pro and the older reasoners top out at high/max, and
-		// OpenRouter's non-flash DeepSeek route exposes only high.
+	if (
+		(isOpenAICompatReasoningApi(spec.api) || (spec.api === "ollama-chat" && spec.provider === "ollama-cloud")) &&
+		isDeepseekReasoningModel(spec)
+	) {
+		// DeepSeek V4 (Flash and Pro) accepts the wire-exact low/high/max ladder
+		// on every first-party/aggregator host — the direct API, aggregators, and
+		// Ollama Cloud alike (medium/xhigh fold into high, max is a real wire
+		// tier). See https://api-docs.deepseek.com/api/create-chat-completion.
+		// OpenRouter's non-Flash V4 route still exposes only high; the older
+		// reasoners (V3.x, R1, deepseek-reasoner) top out at high/max.
 		if (isDeepseekV4FlashModelId(spec.id)) {
 			return LOW_HIGH_MAX_REASONING_EFFORTS;
 		}
+		if (bareModelId(spec.id).toLowerCase().includes("deepseek-v4")) {
+			return isOpenRouterThinkingFormat(compat) ? HIGH_ONLY_REASONING_EFFORTS : LOW_HIGH_MAX_REASONING_EFFORTS;
+		}
 		return isOpenRouterThinkingFormat(compat) ? HIGH_ONLY_REASONING_EFFORTS : HIGH_MAX_REASONING_EFFORTS;
 	}
 	if (spec.provider === "baseten" && isOpenAIGptOssModelId(spec.id)) {