@oh-my-pi/pi-catalog 18.6.3 → 18.7.0

package/CHANGELOG.md

@@ -2,6 +2,23 @@
 
 ## [Unreleased]
 
+## [18.7.0] - 2026-10-06
+
+### Added
+
+- Added Mistral Large 4 with reasoning support, image input, a 1M-token context window, and preview pricing.
+- Added configurable thinking levels from low through maximum for MiniMax-M3.1-Flash-Preview; because the model always reasons, requests that disable thinking use the low level.
+- Added Google Antigravity pricing and model support for Claude Opus 5.5 and Sonnet 5.5.
+
+### Changed
+
+- MiniMax Token Plan providers (`minimax-code` and `minimax-code-cn`) now use MiniMax's recommended Anthropic-compatible API for model requests and login key validation.
+- Google Antigravity now exposes Claude Opus 5.5 and Sonnet 5.5 once each, with selectable low, medium, and high thinking levels.
+
+### Fixed
+
+- Added the correct published pricing for GPT-6 Astra's Ultrafast service tier: a 6× multiplier on the OpenAI API and an 8× multiplier on the Codex card.
+
 ## [18.6.3] - 2026-10-06
 
 ### Added

package/dist/types/compat/types.d.ts

@@ -622,6 +622,8 @@ export interface CompiledProvider {
     id: string;
     /** Preferred model id when no explicit selection is made. */
     defaultModel: string;
+    /** Whether the provider participates in automatic default selection (defaults to true). */
+    automaticDefault?: boolean;
     /** Env vars consulted, in order, for the runtime API-key fallback. */
     envVars?: string[];
     /** The runtime creates a model manager even without a valid API key. */

package/dist/types/types.d.ts

@@ -15,6 +15,22 @@ export type KindApiKind = (typeof KIND_API_KINDS)[number];
 export type WebSearchGrounding = "gemini" | "anthropic" | "codex" | "xai" | "openrouter" | "openai";
 /** Non-chat runner protocols accepted by catalog seeds, outside the chat dispatch union. */
 export declare const RUNNER_APIS: readonly ["local-inference", "web-search", "typesafe", "openrouter-decisions", "openai-images", "openrouter-images", "xai-tts", "openai-speech", "openai-embeddings", "openrouter-rerank", "openrouter-video", "openai-transcriptions"];
+/** Kind each single-purpose runner API serves; `local-inference` hosts several kinds. */
+export declare const RUNNER_API_KINDS: Record<Exclude<(typeof RUNNER_APIS)[number], "local-inference">, ModelKind>;
+/** Kind a runner API serves; `undefined` for chat transports and multi-kind `local-inference`. */
+export declare function runnerApiKind(api: Api): ModelKind | undefined;
+/** Catalog APIs `generate_image` runs through a pi-ai image client; the hosted Responses pair needs a carrier model. */
+export declare const IMAGE_GENERATION_APIS: readonly ["openai-images", "openrouter-images", "google-generative-ai", "google-gemini-cli", "openai-responses", "openai-codex-responses"];
+export type ImageGenerationApi = (typeof IMAGE_GENERATION_APIS)[number];
+/**
+ * Kinds a model on `api` may declare: a runner api serves its own kind; a chat
+ * transport serves `chat` and `tiny`, plus `image` when `generate_image` runs it
+ * (hosted Responses image tool, Gemini image models). `undefined` for
+ * `local-inference`, which hosts several kinds chosen by the model itself.
+ */
+export declare function servedKinds(api: Api): readonly ModelKind[] | undefined;
+/** Whether a model of `kind` can run on `api`. */
+export declare function apiServesKind(api: Api, kind: ModelKind): boolean;
 /** Resolve a model's kind while preserving chat semantics for existing catalog rows. */
 export declare function modelKind(model: Pick<Model, "kind">): ModelKind;
 /** Canonical thinking transport used by a model. */
@@ -1090,6 +1106,12 @@ export interface Model<TApi extends Api = Api> {
     id: string;
     /** Role-specific runner capability; omitted for ordinary chat models. */
     kind?: ModelKind;
+    /**
+     * Verbatim configured kind (models.yml, `modelOverrides`, runtime
+     * registrations). `buildModel` applies it over catalog `kind` rules on every
+     * rebuild, so a configured runner keeps its role.
+     */
+    kindConfig?: ModelKind;
     /** Grounding transport supported by this chat model. */
     webSearch?: WebSearchGrounding;
     /** Cheaper same-provider model to run hosted web search in this model's place (model id or provider/id). */
@@ -1302,9 +1324,9 @@ export interface Model<TApi extends Api = Api> {
     /**
      * Per-service-tier cost multipliers baked from the `service-tier-cost`
      * catalog axis (e.g. `{ priority: 2.5 }`). Absent tiers use the API-generic
-     * defaults.
+     * defaults, and a tier with no published price stays at 1x.
      */
-    serviceTierCost?: Readonly<Partial<Record<"flex" | "priority", number>>>;
+    serviceTierCost?: Readonly<Partial<Record<"flex" | "priority" | "ultrafast", number>>>;
     /**
      * Provider-supplied one-line blurb for this model. Set only when an upstream
      * ships one (Devin's `GetCliModelConfigs`); never synthesized locally.

package/package.json

@@ -1,6 +1,6 @@
 {
 	"name": "@oh-my-pi/pi-catalog",
-	"version": "18.6.3",
+	"version": "18.7.0",
 	"description": "Model catalog for omp: bundled model database, provider discovery descriptors, model identity, classification, and equivalence",
 	"keywords": [
 		"ai",
@@ -93,12 +93,12 @@
 		"gen:proto": "bun scripts/generate-protocols.ts"
 	},
 	"dependencies": {
-		"@oh-my-pi/omptype": "18.6.3",
-		"@oh-my-pi/pi-utils": "18.6.3"
+		"@oh-my-pi/omptype": "18.7.0",
+		"@oh-my-pi/pi-utils": "18.7.0"
 	},
 	"devDependencies": {
 		"@bgotink/kdl": "0.4.0",
-		"@oh-my-pi/pi-ai": "18.6.3",
+		"@oh-my-pi/pi-ai": "18.7.0",
 		"@types/bun": "^1.3.14"
 	},
 	"engines": {

package/src/build.ts

@@ -101,9 +101,11 @@ function applyCatalogAssignments<TApi extends Api>(model: Model<TApi>, catalog:
 	if (serviceTierCost !== undefined) {
 		const flex = numberField(serviceTierCost, "flex");
 		const priorityTier = numberField(serviceTierCost, "priority");
+		const ultrafast = numberField(serviceTierCost, "ultrafast");
 		model.serviceTierCost = {
 			...(flex !== undefined && { flex }),
 			...(priorityTier !== undefined && { priority: priorityTier }),
+			...(ultrafast !== undefined && { ultrafast }),
 		};
 	}
 	const priority = catalog.priority;
@@ -363,7 +365,8 @@ export function buildModel<TApi extends Api>(spec: ModelSpec<TApi>): Model<TApi>
 	};
 	applyCatalogAssignments(model, policy.catalog);
 	applyCatalogCorrections(model, policy.catalog);
-	// Configured lifetimes replace catalog lifetimes rather than merging with them.
+	// Configured kinds and lifetimes replace catalog values rather than merging with them.
+	if (spec.kindConfig !== undefined) model.kind = spec.kindConfig;
 	if (spec.promptCacheConfig !== undefined) model.promptCache = { ...spec.promptCacheConfig };
 	return model;
 }

package/src/compat/rules/README.md

@@ -413,6 +413,10 @@ provider "muse-code" {
 
 Only `discovery` enrolls a provider in `generate-models.ts`; providers without it are never fetched at generation time (see the `charm-hyper` entry for why a live gateway deliberately omits it).
 
+`automatic-default #false` keeps a provider available for explicit selection but excludes it from startup fallback and automatic model presets. The default is `#true`; `apple` opts out because its on-device context cannot accommodate the standard coding-agent prompt in many projects.
+
+`kind-apis { <kind> "<api>" }` maps each non-chat kind (`image`, `tts`, `stt`, `embedding`, `rerank`, `video`) to the API discovery assigns rows of that kind. A runner API must sit under the kind it serves (`RUNNER_API_KINDS` in `src/types.ts`); chat APIs, which serve hosted image generation, and multi-kind `local-inference` may back any kind.
+
 ### Seed rows
 
 A `seed` _defines_ bundled rows for providers whose catalog cannot be discovered at generation time — credential-scoped rosters, unauthenticated regens, or models ahead of upstream catalogs. Every other stratum patches rows; this one authors them. Runtime model managers hand the rows to `staticModels` through `seedModels(provider)`; the generator bundles them per the seed's `bundle` policy. Values are literal — a seed never derives from another provider's row, and pricing is never borrowed.

package/src/compat/rules/auth/minimax-code-cn.kdl

@@ -4,6 +4,6 @@ auth "minimax-code-cn" {
 		auth-url "https://platform.minimaxi.com/subscribe/token-plan"
 		instructions "Subscribe to Token Plan and copy your API key"
 		prompt "Paste your MiniMax Token Plan API key" placeholder="sk-..."
-		validate "chat-completions" base-url="https://api.minimaxi.com/v1" model="MiniMax-M3"
+		validate "anthropic-messages" base-url="https://api.minimaxi.com/anthropic" model="MiniMax-M3"
 	}
 }

package/src/compat/rules/auth/minimax-code.kdl

@@ -4,6 +4,6 @@ auth "minimax-code" {
 		auth-url "https://platform.minimax.io/subscribe/token-plan"
 		instructions "Subscribe to Token Plan and copy your API key"
 		prompt "Paste your MiniMax Token Plan API key" placeholder="sk-..."
-		validate "chat-completions" label="MiniMax Token Plan" base-url="https://api.minimax.io/v1" model="MiniMax-M3"
+		validate "anthropic-messages" label="MiniMax Token Plan" base-url="https://api.minimax.io/anthropic" model="MiniMax-M3"
 	}
 }

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

@@ -51,7 +51,7 @@ class "minimax" {
 			thinking-efforts "minimal" "low" "medium" "high" "xhigh"
 		}
 	}
-	on "minimax" "minimax-cn" "vercel-ai-gateway" {
+	on "minimax" "minimax-cn" "minimax-code" "minimax-code-cn" "vercel-ai-gateway" {
 		thinking-mode "anthropic-adaptive"
 		family "m2" {
 			// Replaces the Anthropic-route MiniMax effort remap table.
@@ -71,6 +71,26 @@ class "minimax" {
 			thinking-efforts "low" "medium" "high"
 		}
 	}
+	// MiniMax-M3.1-Flash-Preview always thinks (`thinking.type: "disabled"` and
+	// effort `none` 400 with error 2013) and tunes depth through
+	// `output_config.effort` from `low` to `max` (the default when omitted).
+	// The identity map replaces the M3 family's adaptive-tag-only map so each
+	// level reaches the wire. Like the M2 family, thinking cannot be turned off:
+	// `thinking-requires-effort` hides the off level and clamps forced-off
+	// requests to the lowest effort instead of sending `disabled`.
+	on "minimax" "minimax-cn" "minimax-code" "minimax-code-cn" {
+		models "MiniMax-M3.1-Flash-Preview" {
+			thinking-requires-effort #true
+			thinking-efforts "low" "medium" "high" "xhigh" "max"
+			thinking-effort-map {
+				low "low"
+				medium "medium"
+				high "high"
+			}
+		}
+	}
+	// Token Plan rows ride the Anthropic route; these OpenAI-only wire axes
+	// serve the legacy lightning rows that stay on the OpenAI-compatible one.
 	on "minimax-code" "minimax-code-cn" {
 		supports-reasoning-effort #false
 		supports-store #false

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

@@ -5,4 +5,5 @@
 provider "apple" {
 	default-model "on-device"
 	allow-unauthenticated #true
+	automatic-default #false
 }

package/src/compat/rules/providers/minimax-code-cn.kdl

@@ -6,11 +6,16 @@ provider "minimax-code-cn" priority=1 {
 
 	// Replaces the provider-id regex in both OpenAI baselines.
 	reasoning-deltas-may-be-cumulative #true
-	class "minimax" {
+	// residue: the legacy lightning SKUs survive only as previous-snapshot rows
+	// on the OpenAI-compatible route (models.dev and both live model lists omit
+	// them), so they keep that route's effort-mode thinking instead of the
+	// Anthropic adaptive rules the rest of the plan uses.
+	models "MiniMax-M2.1-lightning" "MiniMax-M2.5-lightning" {
 		thinking-mode "effort"
-	}
-	// residue: taxonomy ranks and exact globs do not isolate these models.
-	models "MiniMax-M3" {
-		thinking-efforts "minimal" "low" "medium" "high"
+		thinking-effort-map {
+			low "low"
+			medium "medium"
+			high "high"
+		}
 	}
 }

package/src/compat/rules/providers/minimax-code.kdl

@@ -6,11 +6,16 @@ provider "minimax-code" priority=1 {
 
 	// Replaces the provider-id regex in both OpenAI baselines.
 	reasoning-deltas-may-be-cumulative #true
-	class "minimax" {
+	// residue: the legacy lightning SKUs survive only as previous-snapshot rows
+	// on the OpenAI-compatible route (models.dev and both live model lists omit
+	// them), so they keep that route's effort-mode thinking instead of the
+	// Anthropic adaptive rules the rest of the plan uses.
+	models "MiniMax-M2.1-lightning" "MiniMax-M2.5-lightning" {
 		thinking-mode "effort"
-	}
-	// residue: taxonomy ranks and exact globs do not isolate these models.
-	models "MiniMax-M3" {
-		thinking-efforts "minimal" "low" "medium" "high"
+		thinking-effort-map {
+			low "low"
+			medium "medium"
+			high "high"
+		}
 	}
 }

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

@@ -4,6 +4,26 @@ provider "mistral" {
 	default-model "devstral-medium-latest"
 	env "MISTRAL_API_KEY"
 
+	// Public-preview aliases and current USD / 1M token pricing:
+	// https://docs.mistral.ai/models/mistral-large-4-0 (2026-10-06).
+	// Public preview is 50% off: input $0.68, output $2.09, cached input $0.07.
+	// List prices: input $1.36, output $4.18, cached input $0.14; update when preview ends.
+	// No separate maximum output-token limit is published.
+	seed api="openai-completions" base-url="https://api.mistral.ai/v1" bundle="always" {
+		model "mistral-large-4" name="Mistral Large 4" {
+			reasoning #true
+			input "text" "image"
+			cost input=0.68 output=2.09 cache-read=0.07 cache-write=0
+			limits context=1000000
+		}
+		model "mistral-large-4-0" name="Mistral Large 4" {
+			reasoning #true
+			input "text" "image"
+			cost input=0.68 output=2.09 cache-read=0.07 cache-write=0
+			limits context=1000000
+		}
+	}
+
 	thinking-mode "effort"
 	class "unknown" {
 		thinking-efforts "minimal" "low" "medium" "high" "xhigh"

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

@@ -76,6 +76,12 @@ provider "openai-codex" {
 		service-tier-cost {
 			flex 0.5
 			priority 2.5
+			// Ultrafast is Astra-only. This table uses OpenAI's included-usage
+			// multipliers for Codex: Ultrafast 8x and Fast 2.5x Standard (purchased
+			// credits / Enterprise pay-as-you-go bill 6x and 2x instead). See
+			// https://learn.chatgpt.com/docs/agent-configuration/speed and
+			// https://learn.chatgpt.com/docs/pricing.
+			ultrafast 8
 		}
 		limits-patch {
 			context-window 272000

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

@@ -176,6 +176,11 @@ provider "openai" {
 			cache-read 2.0
 			cache-write 25.0
 		}
+		// Ultrafast is Astra-only: $60/$6/$75/$300 per 1M short-context
+		// (6x the standard card), and the same 6x holds on the long-context tier.
+		service-tier-cost {
+			ultrafast 6
+		}
 	}
 	// residue: the bare Daybreak rolling aliases pin the documented 5.6-gen
 	// wire restrictions without carrying the gpt- prefix identity.

package/src/compat/rules/runtime/behavior.kdl

@@ -351,6 +351,8 @@ behavior {
 		alias "claude-opus-4-6" peer-id="claude-opus-4-6@default"
 		alias "claude-sonnet-4-5" peer-id="claude-sonnet-4-5@20250929"
 		alias "claude-sonnet-4-6" peer-id="claude-sonnet-4-6@default"
+		alias "claude-opus-5-5" peer-id="claude-opus-5-5@default"
+		alias "claude-sonnet-5-5" peer-id="claude-sonnet-5-5@default"
 	}
 	// SuperGrok surfaces a few models under IDs that differ from their public
 	// xai equivalent, so the exact-id price mirror misses them.

package/src/compat/rules/taxonomy/_collapse.kdl

@@ -88,6 +88,24 @@ collapse {
 		mode "budget"
 		efforts "minimal" "low" "medium" "high"
 	}
+	variant-family "google-antigravity" "claude-opus-5-5" name="Claude Opus 5.5" {
+		members "claude-opus-5-5-low" "claude-opus-5-5-medium" "claude-opus-5-5-high"
+		route "low" "claude-opus-5-5-low"
+		route "medium" "claude-opus-5-5-medium"
+		route "high" "claude-opus-5-5-high"
+		mode "budget"
+		efforts "low" "medium" "high"
+		requires-effort #true
+	}
+	variant-family "google-antigravity" "claude-sonnet-5-5" name="Claude Sonnet 5.5" {
+		members "claude-sonnet-5-5-low" "claude-sonnet-5-5-medium" "claude-sonnet-5-5-high"
+		route "low" "claude-sonnet-5-5-low"
+		route "medium" "claude-sonnet-5-5-medium"
+		route "high" "claude-sonnet-5-5-high"
+		mode "budget"
+		efforts "low" "medium" "high"
+		requires-effort #true
+	}
 	variant-family "google-antigravity" "claude-sonnet-4-6" name="Claude Sonnet 4.6" {
 		members "claude-sonnet-4-6" "claude-sonnet-4-6-thinking"
 		mode "budget"