@oh-my-pi/pi-catalog 15.11.6 → 15.11.8

package/CHANGELOG.md

@@ -2,6 +2,32 @@
 
 ## [Unreleased]
 
+## [15.11.8] - 2026-06-12
+
+### Fixed
+
+- Fixed Antigravity `gemini-3.1-pro --thinking high` failing with `Cloud Code Assist API error (400): Request contains an invalid argument.` — the upstream `gemini-3.1-pro-high` deployment rejects every `streamGenerateContent` request on both CCA endpoints while discovery still advertises it. High effort now routes to `gemini-pro-agent` (the same "Gemini 3.1 Pro (High)" model, verified accepting the identical request body), and the model-cache fingerprint version was bumped (`merge-v2` → `merge-v3`) so existing fresh caches refetch discovery and pick up the corrected routing immediately.
+
+## [15.11.7] - 2026-06-12
+### Added
+
+- Added effort-tier variant collapsing (`variant-collapse`): providers that expose one logical model as several effort/thinking-suffixed upstream ids (Antigravity CCA `gemini-3.5-flash-extra-low`/`-low`/`gemini-3-flash-agent`, `gemini-3[.1]-pro-low|high`, `claude-*[-thinking]` pairs, `gpt-oss-120b-medium`) collapse into one logical entry carrying per-effort upstream routing in `thinking.effortRouting` (plus `thinking.suppressWhenOff` for Cloud Code Assist ids whose baked server default re-applies when `thinkingConfig` is omitted). Request-time code resolves the outbound id via `resolveWireModelId(model, effort)`; selection, caching, and usage attribution key on the logical id.
+- Added the automatic `X`/`X-thinking` pair rule (`deriveThinkingPairFamilies`): any provider's live bare/thinking twin collapses into the bare id, routing thinking-enabled requests to the `-thinking` backing id (trailing or infix token, so `kimi-k2-thinking-turbo` pairs with `kimi-k2-turbo`). Gated on same api and compatible pricing — all-zero cost rows count as unknown, while twins that both carry real, differing prices remain separate SKUs.
+- Added `collapseBuiltModelVariants` and wired collapsing at every materialization point — Antigravity discovery, the catalog generator, and the model-manager merge — so stale sources (old static beside collapsed dynamic results, mixed cache rows) converge on logical entries instead of unioning raw tier ids back into the catalog.
+- Added `thinking.requiresEffort`, baked for reasoning-only upstreams — Gemini 3.x (levels only, no off), Gemini 2.5 Pro (thinkingBudget floors at 128, rejects 0), OpenAI o-series, MiniMax M2, and thinking-variant SKUs (`*-thinking`/`*-reasoner`/`*-reasoning`, with a negation-aware token grammar so `non-thinking` ids never match). Identity derivation bakes it for new entries and `fillThinkingWireDefaults` backfills explicit/cached metadata; `minimumSupportedEffort` exposes the canonical floor. Pair-collapsed twins drop member flags (their off routes to the bare SKU), while identity re-flags pairs whose logical id is itself mandatory
+
+### Changed
+
+- Changed model display names to drop model-extrinsic decorations: gateway author prefixes (`OpenAI: …`, `Google: …`), `(latest)` alias markers, `(Antigravity)` provider attribution, price tiers (`($$$$)`), and promo/lifecycle tags (`(20% off)`, `(retires …)`). `cleanModelName` is applied in `buildModel` (covers live discovery and stale caches) and as a catalog-generator pass; Antigravity discovery no longer appends `(Antigravity)` to display names. Variant tags that map to distinct wire ids (`(Thinking)`, `(free)`, `(Fast)`, dates, regions) are preserved.
+- Changed the `google-antigravity` default model from `gemini-3-pro-high` to `gemini-3.1-pro`
+- Changed `gemini-2.5-flash-thinking` handling from discovery-denylist to collapsing into `gemini-2.5-flash` (thinking-enabled requests route to the `-thinking` backing id)
+- Bumped the model cache schema to v5 so rows predating effort-tier variant collapsing (raw `-low`/`-high`/`-thinking` member ids) are invalidated
+
+### Fixed
+
+- Fixed catalog generation to apply effort-tier variant collapsing before provider grouping to ensure collapsed model families are consistently materialized without being impacted by in-loop mutation
+- Fixed Kimi K2.6 OpenAI-compatible compat metadata to use a 300s stream watchdog floor, covering Fire Pass router ids as well as public `kimi-k2.6` ids so long reasoning starts do not hit the generic first-event timeout ([#2366](https://github.com/can1357/oh-my-pi/issues/2366)).
+
 ## [15.11.4] - 2026-06-12
 
 ### Fixed

package/README.md

@@ -0,0 +1,43 @@
+# @oh-my-pi/pi-catalog
+
+Model catalog for [oh-my-pi](https://github.com/can1357/oh-my-pi): bundled model database, provider discovery, model identity, classification, and equivalence.
+
+## What's inside
+
+| Module | Purpose |
+| --- | --- |
+| `models.json` + `models` | Bundled model database (pricing, context windows, modalities, thinking support) |
+| `provider-models` | Provider catalog descriptors (`CATALOG_PROVIDERS`), per-provider model resolution rules |
+| `discovery` | Runtime model discovery for OpenAI-compatible endpoints, Gemini, Codex, Cursor, Antigravity, Ollama |
+| `identity` | Model id parsing and classification (family/version), reference resolution, equivalence, selection priority |
+| `model-thinking` | Thinking/reasoning metadata and generated per-model policies |
+| `model-manager` / `model-cache` | Runtime model registry with discovery refresh and on-disk caching |
+| `variant-collapse` | Collapsing provider-specific variants of the same underlying model |
+| `compat` | Request/response compatibility fixups for OpenAI- and Anthropic-shaped APIs |
+| `wire` | Wire-level helpers: Codex, Gemini headers, GitHub Copilot |
+| `effort` | Reasoning-effort level definitions |
+
+Import from subpaths (`@oh-my-pi/pi-catalog/<module>`) or the root barrel.
+
+## models.json is generated
+
+Never edit `src/models.json` by hand — it is produced from upstream sources (models.dev, provider catalog discovery, OpenCode docs) by `scripts/generate-models.ts` and the resolvers in `src/provider-models/`. Regenerate with:
+
+```sh
+bun --cwd=packages/catalog run generate-models
+```
+
+To change an entry, fix the source: resolver overrides in `provider-models/openai-compat.ts`, provider entries in `provider-models/descriptors.ts`, generator fixups in `scripts/generate-models.ts`, or thinking policies in `model-thinking.ts`.
+
+## Install
+
+```sh
+bun add @oh-my-pi/pi-catalog
+```
+
+Ships TypeScript source directly (no build step); requires Bun ≥ 1.3.14.
+
+## References
+
+- [Monorepo README](https://github.com/can1357/oh-my-pi#readme)
+- [CHANGELOG](./CHANGELOG.md)

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

@@ -8,7 +8,7 @@
  */
 /** Kimi family ids in any namespace form (`moonshotai/kimi-*`, `kimi-k2.6`, `vendor/kimi.x`). */
 export declare function isKimiModelId(modelId: string): boolean;
-/** Kimi K2.6 specifically (preserved-thinking transport on Moonshot-native hosts). */
+/** Kimi K2.6 specifically, including router ids that spell the version `k2p6`. */
 export declare function isKimiK26ModelId(modelId: string): boolean;
 /** Claude ids in any namespace form (`claude-*`, `vendor/claude.x`). */
 export declare function isClaudeModelId(modelId: string): boolean;
@@ -60,3 +60,23 @@ export declare function hasOpus47ApiRestrictions(modelId: string): boolean;
  */
 export declare function supportsMidConversationSystemMessages(modelId: string): boolean;
 export declare function isAnthropicFableOrMythosModel(modelId: string): boolean;
+/** Thinking-variant token location inside a model id. */
+export interface ThinkingVariantToken {
+    index: number;
+    length: number;
+}
+/**
+ * Locates the first thinking-variant token (`-thinking`, `-reasoner`,
+ * `-reasoning`; trailing or infix) in a model id. The token ends at the id
+ * end or any non-alphanumeric boundary, and negated forms (`non-thinking`,
+ * `no-thinking`) never match — those name the NON-thinking SKU.
+ */
+export declare function findThinkingVariantToken(modelId: string): ThinkingVariantToken | undefined;
+/**
+ * Removes the located thinking-variant token: `kimi-k2-thinking` → `kimi-k2`,
+ * `mimo-v2-flash-thinking-original` → `mimo-v2-flash-original`,
+ * `grok-4.1-fast-reasoning` → `grok-4.1-fast`. Returns `undefined` when no
+ * token exists or nothing would remain. Callers MUST verify the result names
+ * a live model.
+ */
+export declare function stripThinkingVariantToken(modelId: string): string | undefined;

package/dist/types/index.d.ts

@@ -10,6 +10,7 @@ export * from "./models";
 export * from "./provider-models";
 export * from "./types";
 export * from "./utils";
+export * from "./variant-collapse";
 export * from "./wire/codex";
 export * from "./wire/gemini-headers";
 export * from "./wire/github-copilot";

package/dist/types/model-thinking.d.ts

@@ -64,4 +64,16 @@ export declare function mapEffortToGoogleThinkingLevel(effort: Effort): "MINIMAL
  * the model's baked `thinking.effortMap` (identity for unmapped efforts).
  */
 export declare function mapEffortToAnthropicAdaptiveEffort<TApi extends Api>(model: ApiModel<TApi>, effort: Effort): "low" | "medium" | "high" | "xhigh" | "max";
+/**
+ * Resolves the upstream wire model id for a request at the given effort
+ * (`undefined` = thinking off). Collapsed effort-tier variants route through
+ * `thinking.effortRouting`; everything else falls back to
+ * `requestModelId ?? id`.
+ */
+export declare function resolveWireModelId<TApi extends Api>(model: ApiModel<TApi>, effort: Effort | undefined): string;
+/**
+ * Lowest supported effort in canonical order — the clamp target for
+ * thinking-off requests on `thinking.requiresEffort` models.
+ */
+export declare function minimumSupportedEffort<TApi extends Api>(model: ApiModel<TApi>): Effort | undefined;
 export {};

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

@@ -93,7 +93,7 @@ export declare const CATALOG_PROVIDERS: readonly [{
     readonly createModelManagerOptions: (config: ModelManagerConfig) => import("..").ModelManagerOptions<"google-generative-ai", unknown>;
 }, {
     readonly id: "google-antigravity";
-    readonly defaultModel: "gemini-3-pro-high";
+    readonly defaultModel: "gemini-3.1-pro";
     readonly specialModelManager: true;
 }, {
     readonly id: "google-gemini-cli";

package/dist/types/types.d.ts

@@ -27,6 +27,28 @@ export interface ThinkingConfig {
      * 5). Also implies native interleaved thinking — no beta header needed.
      */
     supportsDisplay?: boolean;
+    /**
+     * Per-effort upstream wire-id routing for collapsed effort-tier variants
+     * (`variant-collapse.ts`). Keyed by pi effort; `"off"` applies when
+     * thinking is disabled. Missing keys fall back to `requestModelId ?? id`.
+     */
+    effortRouting?: Readonly<Partial<Record<Effort | "off", string>>>;
+    /**
+     * When true, a thinking-off request MUST explicitly suppress thinking on
+     * the wire (google-level: `thinkingLevel: "MINIMAL"` + `includeThoughts:
+     * false`; budget: `thinkingBudget: 0`) instead of omitting thinkingConfig —
+     * Cloud Code Assist re-applies the per-id baked server default when the
+     * config is absent.
+     */
+    suppressWhenOff?: boolean;
+    /**
+     * Reasoning is mandatory upstream: the endpoint rejects disabled or
+     * omitted thinking (e.g. OpenRouter Gemini 3.x — "Reasoning is mandatory
+     * for this endpoint and cannot be disabled"). Request mapping clamps
+     * thinking-off to the lowest supported effort unless `suppressWhenOff`
+     * provides an explicit wire off-path.
+     */
+    requiresEffort?: boolean;
 }
 export type Provider = string;
 /** Token budgets for each thinking level (token-based providers only) */

package/dist/types/utils.d.ts

@@ -3,3 +3,9 @@ export declare function toNumber(value: unknown): number | undefined;
 export declare function toPositiveNumber(value: unknown, fallback: number): number;
 export declare function toBoolean(value: unknown): boolean | undefined;
 export declare function isAnthropicOAuthToken(key: string): boolean;
+/**
+ * Normalize a model display name: drop the gateway author prefix and
+ * model-extrinsic decorations. Returns the input verbatim when nothing
+ * matches (or when stripping would leave an empty name).
+ */
+export declare function cleanModelName(name: string): string;

package/dist/types/variant-collapse.d.ts

@@ -0,0 +1,126 @@
+import { Effort } from "./effort";
+import type { Api, Model, ModelSpec, Provider, ThinkingConfig } from "./types";
+/**
+ * Structural bound for collapse inputs: both raw `ModelSpec`s and built
+ * `Model`s qualify. (`Model.compat` is the resolved record, not the sparse
+ * config, so the two are not mutually assignable — collapsing never touches
+ * `compat`.)
+ */
+export type VariantSpecLike = Omit<ModelSpec<Api>, "compat"> & {
+    compat?: unknown;
+};
+/** One collapsed family: logical id + member wire ids + per-effort routing. */
+export interface EffortVariantFamily {
+    /** Collapsed logical id (may equal a member id — e.g. bare/thinking pairs). */
+    id: string;
+    /** Final display name, no tier marker. */
+    name: string;
+    /**
+     * Member wire ids in priority order. The first member present in the input
+     * becomes the collapsed spec's default wire id (`requestModelId`; omitted
+     * when it equals the logical id).
+     */
+    members: readonly string[];
+    /**
+     * Wire ids upstream no longer serves (e.g. a deployment killed while
+     * discovery still advertises it). Fresh collapsing never routes to them,
+     * and stale collapsed snapshots (bundled catalog, cache rows,
+     * previous-generation fallbacks) get routing/`requestModelId` entries that
+     * target them re-pointed through `routing`. Keep retired ids in `members`
+     * so the raw upstream spec is still consumed and aliased.
+     */
+    retiredMembers?: readonly string[];
+    /**
+     * Per-effort upstream wire id; `"off"` applies when thinking is disabled.
+     * Entries whose target member is absent from the input are dropped — those
+     * efforts fall back to `requestModelId ?? id`.
+     */
+    routing: Readonly<Partial<Record<Effort | "off", string>>>;
+    /** Explicit capability surface for the collapsed spec — no inference. */
+    thinking: Readonly<Omit<ThinkingConfig, "effortRouting" | "suppressWhenOff">>;
+    /** Thinking-off requests must explicitly suppress thinking on the wire. */
+    suppressWhenOff?: boolean;
+    /** Retired/recycled selector ids that alias to this family without being members. */
+    extraAliases?: readonly string[];
+}
+export interface VariantCollapseTable {
+    families: readonly EffortVariantFamily[];
+}
+/**
+ * Shared by `google-antigravity` and `google-gemini-cli` — both serve the
+ * Antigravity discovery list (`fetchAntigravityDiscoveryModels`).
+ */
+export declare const ANTIGRAVITY_VARIANT_COLLAPSE_TABLE: VariantCollapseTable;
+/** Provider id → hand collapse table. Both CCA providers share one table. */
+export declare const VARIANT_COLLAPSE_TABLES: Readonly<Record<string, VariantCollapseTable>>;
+/**
+ * The global automatic rule: derive an `X` + `X-thinking` family for every
+ * pair where both ids are live in `specs` (trailing or infix token). Gates:
+ * - both members share the same `api`,
+ * - known pricing must match — all-zero cost rows count as unknown
+ *   (aggregators routinely ship them), but twins that BOTH carry real,
+ *   differing prices are distinct SKUs and never merge,
+ * - ids claimed by the provider's hand `table` are skipped (curation wins).
+ * The capability surface prefers the thinking member's metadata, then the
+ * bare member's, then the canonical deriver (aggregators often ship
+ * `reasoning: false` and no thinking config on the twin), then a budget
+ * default. `off` routes to the bare id; every supported effort routes to the
+ * thinking id.
+ */
+export declare function deriveThinkingPairFamilies<TSpec extends VariantSpecLike>(specs: readonly TSpec[], table?: VariantCollapseTable): EffortVariantFamily[];
+/**
+ * True when `spec` is the output of collapsing rather than a raw upstream
+ * member. `thinking.effortRouting` is written only by collapsing; the
+ * `requestModelId` arm is scoped to the provider's hand-table family ids so
+ * unrelated carriers (GitHub Copilot `-1m` context variants) never match.
+ */
+export declare function isVariantCollapsedSpec(spec: VariantSpecLike): boolean;
+/**
+ * Collapse every family in `table` found in `specs`. Non-member specs pass
+ * through verbatim (by reference), order preserved; the collapsed spec
+ * replaces the first occurrence of its family.
+ */
+export declare function collapseEffortVariants<TSpec extends VariantSpecLike>(specs: readonly TSpec[], table: VariantCollapseTable): TSpec[];
+/**
+ * Collapse a full mixed-provider list: per provider, the hand table (when
+ * registered) plus the automatic `X`/`X-thinking` pair rule. Used by the
+ * catalog generator; the runtime equivalent lives at the model-manager merge
+ * point. Output is regrouped by provider — callers re-sort.
+ */
+export declare function collapseEffortVariantsAcrossProviders<TSpec extends VariantSpecLike>(specs: readonly TSpec[]): TSpec[];
+/**
+ * Runtime entry point for already-built `Model` lists (the model-manager
+ * merge point, coding-agent registry custom providers): collapses hand
+ * tables plus derived pairs, then re-runs `buildModel` on freshly created
+ * logical specs so thinking wire defaults stay resolved. Untouched entries
+ * pass through by reference.
+ */
+export declare function collapseBuiltModelVariants<TApi extends Api>(models: readonly Model<TApi>[]): Model<TApi>[];
+/**
+ * Resolve a retired effort-tier variant id (collapsed member, recycled id) to
+ * its replacement model id for `provider` via the hand table. Returns
+ * `undefined` when the id is not a known alias; derived `X-thinking` members
+ * resolve through `stripThinkingVariantToken` instead. Callers must try an
+ * exact model lookup first — a live model always wins over an alias.
+ */
+export declare function resolveVariantAlias(provider: Provider, modelId: string): string | undefined;
+/** Bare-id alias hit: replacement id plus the providers declaring it. */
+export interface BareVariantAliasHit {
+    id: string;
+    /** Providers whose table declares the alias — candidates from these win ties. */
+    providers: readonly Provider[];
+}
+/**
+ * Provider-agnostic hand-table alias lookup for bare-id selectors. Returns
+ * the declaring providers so callers can prefer their models when the
+ * replacement id exists on unrelated providers too (e.g. a retired Cursor
+ * tier id must not resolve to `openai/gpt-5.4`).
+ */
+export declare function resolveBareVariantAlias(modelId: string): BareVariantAliasHit | undefined;
+/**
+ * Reverse alias lookup: the retired ids that resolve to `modelId` for
+ * `provider` via the hand table. Used to re-key config keyed by raw member
+ * ids (models.yml `modelOverrides`, suppressed selectors) onto the collapsed
+ * model. Empty for providers without a table.
+ */
+export declare function getVariantAliasSources(provider: Provider, modelId: string): readonly string[];

package/package.json

@@ -1,7 +1,7 @@
 {
 	"type": "module",
 	"name": "@oh-my-pi/pi-catalog",
-	"version": "15.11.6",
+	"version": "15.11.8",
 	"description": "Model catalog for omp: bundled model database, provider discovery descriptors, model identity, classification, and equivalence",
 	"homepage": "https://omp.sh",
 	"author": "Can Boluk",
@@ -34,11 +34,11 @@
 	},
 	"dependencies": {
 		"@bufbuild/protobuf": "^2.12.0",
-		"@oh-my-pi/pi-utils": "15.11.6",
+		"@oh-my-pi/pi-utils": "15.11.8",
 		"zod": "4.4.3"
 	},
 	"devDependencies": {
-		"@oh-my-pi/pi-ai": "15.11.6",
+		"@oh-my-pi/pi-ai": "15.11.8",
 		"@types/bun": "^1.3.14"
 	},
 	"engines": {

package/src/build.ts

@@ -13,11 +13,13 @@ import { buildAnthropicCompat } from "./compat/anthropic";
 import { buildOpenAICompat, buildOpenAIResponsesCompat } from "./compat/openai";
 import { resolveModelThinking } from "./model-thinking";
 import type { Api, CompatOf, Model, ModelSpec } from "./types";
+import { cleanModelName } from "./utils";
 
 export function buildModel<TApi extends Api>(spec: ModelSpec<TApi>): Model<TApi> {
 	const compat = buildCompat(spec) as CompatOf<TApi>;
 	return {
 		...spec,
+		name: cleanModelName(spec.name),
 		thinking: resolveModelThinking(spec, compat),
 		compat,
 		compatConfig: spec.compat,

package/src/compat/openai.ts

@@ -25,6 +25,8 @@ const GLM_CODING_PLAN_MODEL_PATTERN = /^glm-5(?:[.-]|$)/i;
 const GLM_CODING_PLAN_STREAM_IDLE_TIMEOUT_MS = 600_000;
 /** Direct DeepSeek reasoning models stall between thinking and answer phases. */
 const DEEPSEEK_REASONING_STREAM_IDLE_TIMEOUT_MS = 300_000;
+/** Kimi K2.6 can spend several minutes reasoning before the first visible token. */
+const KIMI_K26_REASONING_STREAM_IDLE_TIMEOUT_MS = 300_000;
 
 /**
  * OpenCode's gateways (https://opencode.ai/zen|go) gate `reasoning_content`
@@ -178,15 +180,17 @@ export function buildOpenAICompat(spec: ModelSpec<"openai-completions">): Resolv
 			isCopilotHost ||
 			isZenmuxHost);
 
-	// Stream-watchdog floor: GLM coding-plan SKUs and direct DeepSeek reasoning
-	// models idle for minutes mid-reasoning; widen the idle timeout so warm-ups
-	// stop aborting and retrying.
+	// Stream-watchdog floor: GLM coding-plan SKUs, Kimi K2.6, and direct
+	// DeepSeek reasoning models can idle for minutes while reasoning; widen the
+	// idle timeout so warm-ups stop aborting and retrying.
 	const streamIdleTimeoutMs =
 		GLM_CODING_PLAN_MODEL_PATTERN.test(spec.id) && (isZai || isZhipu)
 			? GLM_CODING_PLAN_STREAM_IDLE_TIMEOUT_MS
-			: spec.reasoning && isDirectDeepseekApi
-				? DEEPSEEK_REASONING_STREAM_IDLE_TIMEOUT_MS
-				: undefined;
+			: spec.reasoning && isKimiK26ModelId(spec.id)
+				? KIMI_K26_REASONING_STREAM_IDLE_TIMEOUT_MS
+				: spec.reasoning && isDirectDeepseekApi
+					? DEEPSEEK_REASONING_STREAM_IDLE_TIMEOUT_MS
+					: undefined;
 
 	const compat: ResolvedOpenAICompat = {
 		supportsStore: !isNonStandard,

package/src/discovery/antigravity.ts

@@ -1,6 +1,7 @@
 import * as z from "zod/v4";
 import type { ModelSpec } from "../types";
 import { toPositiveNumber } from "../utils";
+import { ANTIGRAVITY_VARIANT_COLLAPSE_TABLE, collapseEffortVariants } from "../variant-collapse";
 import { getAntigravityUserAgent } from "../wire/gemini-headers";
 
 const DEFAULT_ANTIGRAVITY_DISCOVERY_ENDPOINTS = [
@@ -11,13 +12,7 @@ const FETCH_AVAILABLE_MODELS_PATH = "/v1internal:fetchAvailableModels";
 
 const DEFAULT_CONTEXT_WINDOW = 200_000;
 const DEFAULT_MAX_TOKENS = 64_000;
-const ANTIGRAVITY_DISCOVERY_DENYLIST = new Set([
-	"chat_20706",
-	"chat_23310",
-	"gemini-2.5-flash-thinking",
-	"gemini-3-pro-low",
-	"gemini-2.5-pro",
-]);
+const ANTIGRAVITY_DISCOVERY_DENYLIST = new Set(["chat_20706", "chat_23310", "gemini-2.5-pro"]);
 
 /**
  * Raw model metadata returned by Antigravity's `fetchAvailableModels` endpoint.
@@ -224,7 +219,7 @@ export async function fetchAntigravityDiscoveryModels(
 			const supportsImages = model.supportsImages === true;
 			models.push({
 				id: modelId,
-				name: model.displayName ? `${model.displayName} (Antigravity)` : modelId,
+				name: model.displayName || modelId,
 				api: "google-gemini-cli",
 				provider: "google-antigravity",
 				baseUrl: endpoint,
@@ -241,8 +236,12 @@ export async function fetchAntigravityDiscoveryModels(
 			});
 		}
 
-		models.sort((a, b) => a.name.localeCompare(b.name) || a.id.localeCompare(b.id));
-		return models;
+		// Collapse effort-tier variants at the source so runtime discovery,
+		// the gemini-cli re-provision, and the catalog generator all see
+		// logical ids only.
+		const collapsed = collapseEffortVariants(models, ANTIGRAVITY_VARIANT_COLLAPSE_TABLE);
+		collapsed.sort((a, b) => a.name.localeCompare(b.name) || a.id.localeCompare(b.id));
+		return collapsed;
 	}
 
 	return null;

package/src/identity/family.ts

@@ -14,9 +14,9 @@ export function isKimiModelId(modelId: string): boolean {
 	return modelId.includes("moonshotai/kimi") || /(^|\/)kimi[-.]/i.test(modelId);
 }
 
-/** Kimi K2.6 specifically (preserved-thinking transport on Moonshot-native hosts). */
+/** Kimi K2.6 specifically, including router ids that spell the version `k2p6`. */
 export function isKimiK26ModelId(modelId: string): boolean {
-	return /(^|\/)kimi-k2\.6(?:[-:]|$)/i.test(modelId);
+	return /(^|\/)kimi-k2(?:\.6|p6)(?:[-:]|$)/i.test(modelId);
 }
 
 /** Claude ids in any namespace form (`claude-*`, `vendor/claude.x`). */
@@ -113,3 +113,44 @@ export function isAnthropicFableOrMythosModel(modelId: string): boolean {
 	const parsed = parseAnthropicModel(bareModelId(modelId));
 	return parsed !== null && isFableOrMythos(parsed.kind);
 }
+
+/** Thinking-variant token location inside a model id. */
+export interface ThinkingVariantToken {
+	index: number;
+	length: number;
+}
+
+const THINKING_VARIANT_TOKEN_RE = /-(?:thinking|reasoner|reasoning)(?=$|[^a-z0-9])/gi;
+
+/**
+ * Locates the first thinking-variant token (`-thinking`, `-reasoner`,
+ * `-reasoning`; trailing or infix) in a model id. The token ends at the id
+ * end or any non-alphanumeric boundary, and negated forms (`non-thinking`,
+ * `no-thinking`) never match — those name the NON-thinking SKU.
+ */
+export function findThinkingVariantToken(modelId: string): ThinkingVariantToken | undefined {
+	THINKING_VARIANT_TOKEN_RE.lastIndex = 0;
+	let match = THINKING_VARIANT_TOKEN_RE.exec(modelId);
+	while (match !== null) {
+		const preceding = /([a-z0-9]+)$/i.exec(modelId.slice(0, match.index))?.[1]?.toLowerCase();
+		if (preceding !== "non" && preceding !== "no") {
+			return { index: match.index, length: match[0].length };
+		}
+		match = THINKING_VARIANT_TOKEN_RE.exec(modelId);
+	}
+	return undefined;
+}
+
+/**
+ * Removes the located thinking-variant token: `kimi-k2-thinking` → `kimi-k2`,
+ * `mimo-v2-flash-thinking-original` → `mimo-v2-flash-original`,
+ * `grok-4.1-fast-reasoning` → `grok-4.1-fast`. Returns `undefined` when no
+ * token exists or nothing would remain. Callers MUST verify the result names
+ * a live model.
+ */
+export function stripThinkingVariantToken(modelId: string): string | undefined {
+	const token = findThinkingVariantToken(modelId);
+	if (!token) return undefined;
+	const stripped = modelId.slice(0, token.index) + modelId.slice(token.index + token.length);
+	return stripped.length > 0 ? stripped : undefined;
+}

package/src/index.ts

@@ -10,6 +10,7 @@ export * from "./models";
 export * from "./provider-models";
 export * from "./types";
 export * from "./utils";
+export * from "./variant-collapse";
 export * from "./wire/codex";
 export * from "./wire/gemini-headers";
 export * from "./wire/github-copilot";

package/src/model-cache.ts

@@ -7,9 +7,10 @@ import { getModelDbPath } from "@oh-my-pi/pi-utils";
 import type { Api, Model, ModelSpec } from "./types";
 
 // Rows persist ModelSpec JSON (sparse `compat`, never the resolved record);
-// the model manager rebuilds via `buildModel` on load. v4 invalidates rows
-// carrying the pre-efforts ThinkingConfig shape (minLevel/maxLevel/levels).
-const CACHE_SCHEMA_VERSION = 4;
+// the model manager rebuilds via `buildModel` on load. v5 invalidates rows
+// predating effort-tier variant collapsing (raw `-low`/`-high`/`-thinking`
+// member ids); v4 dropped the pre-efforts ThinkingConfig shape.
+const CACHE_SCHEMA_VERSION = 5;
 
 interface CacheRow {
 	provider_id: string;

package/src/model-manager.ts

@@ -3,6 +3,7 @@ import { readModelCache, writeModelCache } from "./model-cache";
 import { type GeneratedProvider, getBundledModels } from "./models";
 import type { Api, Model, ModelSpec, Provider } from "./types";
 import { isRecord } from "./utils";
+import { collapseBuiltModelVariants } from "./variant-collapse";
 
 const DEFAULT_CACHE_TTL_MS = 2 * 60 * 60 * 1000;
 const NON_AUTHORITATIVE_RETRY_MS = 5 * 60 * 1000;
@@ -134,7 +135,7 @@ export async function resolveProviderModels<TApi extends Api = Api, TModelsDevPa
 	// Re-running `mergeDynamicModels(static, cache)` would just rebuild the same
 	// objects (~800ms in the steady-state cold-start profile for `omp -p hi`).
 	if (!shouldFetchFromNetwork && cache?.fresh && hasAuthoritativeCache && cacheFingerprintMatches) {
-		return { models: passModelList<TApi>(cache.models), stale: false };
+		return { models: collapseBuiltModelVariants(passModelList<TApi>(cache.models)), stale: false };
 	}
 
 	const [fetchedModelsDevModels, fetchedDynamicModels] = shouldFetchFromNetwork
@@ -148,8 +149,9 @@ export async function resolveProviderModels<TApi extends Api = Api, TModelsDevPa
 	const dynamicModels = fetchedDynamicModels ?? [];
 	const mergedWithCache = mergeDynamicModels(mergeModelSources(staticModels, modelsDevModels), cacheModels);
 	const mergedModels = mergeDynamicModels(mergedWithCache, dynamicModels);
-	const models =
-		dynamicModelsAuthoritative && dynamicFetchSucceeded ? retainModelIds(mergedModels, dynamicModels) : mergedModels;
+	const models = collapseBuiltModelVariants(
+		dynamicModelsAuthoritative && dynamicFetchSucceeded ? retainModelIds(mergedModels, dynamicModels) : mergedModels,
+	);
 	const dynamicAuthoritative = !hasDynamicFetcher || dynamicFetchSucceeded || shouldUseFreshCacheAsAuthoritative;
 	if (shouldFetchFromNetwork) {
 		if (dynamicFetchSucceeded) {
@@ -157,7 +159,14 @@ export async function resolveProviderModels<TApi extends Api = Api, TModelsDevPa
 			const snapshotModels = dynamicModelsAuthoritative
 				? retainModelIds(mergedSnapshot, dynamicModels)
 				: mergedSnapshot;
-			writeModelCache(options.providerId, now(), snapshotModels, true, staticFingerprint, dbPath);
+			writeModelCache(
+				options.providerId,
+				now(),
+				collapseBuiltModelVariants(snapshotModels),
+				true,
+				staticFingerprint,
+				dbPath,
+			);
 		} else {
 			// Dynamic fetch failed — update cache with a non-authoritative snapshot so
 			// stale state remains visible while retry backoff still applies.
@@ -165,9 +174,11 @@ export async function resolveProviderModels<TApi extends Api = Api, TModelsDevPa
 			writeModelCache(
 				options.providerId,
 				now(),
-				mergeDynamicModels(
-					mergeModelSources(staticModels, modelsDevModels),
-					normalizeModelList<TApi>(latestCache?.models ?? cache?.models ?? []),
+				collapseBuiltModelVariants(
+					mergeDynamicModels(
+						mergeModelSources(staticModels, modelsDevModels),
+						normalizeModelList<TApi>(latestCache?.models ?? cache?.models ?? []),
+					),
 				),
 				false,
 				staticFingerprint,
@@ -290,7 +301,7 @@ function retainModelIds<TApi extends Api>(
  * arms calling `resolveProviderModels` with the same `staticModels` array)
  * skip the JSON+hash work after the first call.
  */
-const MODEL_CACHE_FINGERPRINT_VERSION = "merge-v2";
+const MODEL_CACHE_FINGERPRINT_VERSION = "merge-v3";
 const kStaticFingerprint = Symbol("model-manager.staticFingerprint");
 type ModelArrayWithFingerprint = readonly Model<Api>[] & { [kStaticFingerprint]?: string };
 function fingerprintStatic<TApi extends Api>(