@oh-my-pi/pi-catalog 18.1.1 → 18.1.3

package/CHANGELOG.md

@@ -2,6 +2,36 @@
 
 ## [Unreleased]
 
+## [18.1.3] - 2026-09-02
+
+### Added
+
+- Added support for Claude Fable 5.1
+
+### Changed
+
+- Updated pricing and context limits for various Claude models
+
+### Fixed
+
+- Claude Sonnet 5 no longer advertises unsupported mid-conversation system messages.
+- Custom GLM 5.2 models on `alibaba-coding-plan` (and other blanket-GLM hosts) no longer crash startup with `AmbiguousOverlapError` ([#10553](https://github.com/can1357/oh-my-pi/issues/10553)).
+- Gemini 3.7 Flash no longer offers the `minimal` thinking effort on direct google-level hosts (`google`, `google-vertex`, `opencode-zen`), which reject `thinkingLevel: MINIMAL` with a 400; budget and reasoning-effort resellers keep the tier ([#10543](https://github.com/can1357/oh-my-pi/issues/10543)).
+- Fixed Alibaba Token Plan discovery for `qwen3.8-flash` to include its context limits, reasoning support, and image input.
+- Z.AI GLM-5.3-Flash now uses the native API instead of failing through the unsupported Anthropic-compatible route ([#10539](https://github.com/can1357/oh-my-pi/issues/10539)).
+
+## [18.1.2] - 2026-09-01
+
+### Added
+
+- Added support for turn-scoped system messages, changing tools during a conversation, and setting effort on individual messages.
+- Added support for Claude Fable 5.1 models, with improved thinking-prefix handling across supported Claude models.
+- Added DeepSeek V4 Flash Vision Exp, Mercury 2.5 Preview, and Xiaomi MiMo V2.5 Pro UltraSpeed models.
+
+### Changed
+
+- Updated model pricing and context-window limits.
+
 ## [18.1.0] - 2026-09-01
 
 ### Added

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

@@ -502,7 +502,7 @@ export declare const CATALOG_PROVIDERS: readonly [{
     readonly id: "zai";
     readonly defaultModel: "glm-5.3";
     readonly envVars: readonly ["ZAI_API_KEY"];
-    readonly createModelManagerOptions: (config: ModelManagerConfig) => import("../index.js").ModelManagerOptions<"anthropic-messages", unknown>;
+    readonly createModelManagerOptions: (config: ModelManagerConfig) => import("../index.js").ModelManagerOptions<"anthropic-messages" | "openai-completions", unknown>;
     readonly catalogDiscovery: {
         readonly label: "zAI";
     };

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

@@ -673,6 +673,11 @@ export interface ModelsDevProviderDescriptor {
 }
 /** Generic mapper that converts models.dev data using provider descriptors. */
 export declare function mapModelsDevToModels(data: Record<string, unknown>, descriptors: readonly ModelsDevProviderDescriptor[]): ModelSpec<Api>[];
+/** Resolves the transport and endpoint for one Z.AI model catalog entry. */
+export declare function resolveZaiApi(modelId: string): {
+    api: "anthropic-messages" | "openai-completions";
+    baseUrl: string;
+};
 /** All provider descriptors for models.dev data mapping in generate-models.ts. */
 export declare const MODELS_DEV_PROVIDER_DESCRIPTORS: readonly ModelsDevProviderDescriptor[];
 /** Providers whose bundled catalog can receive additive models.dev updates at runtime. */

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

@@ -70,4 +70,5 @@ export declare const DEVIN_STATIC_MODELS: readonly ModelSpec<"devin-agent">[];
 export declare function devinModelManagerOptions(config?: DevinModelManagerConfig): ModelManagerOptions<"devin-agent">;
 export interface ZaiModelManagerConfig {
 }
-export declare function zaiModelManagerOptions(_config?: ZaiModelManagerConfig): ModelManagerOptions<"anthropic-messages">;
+/** Creates model-manager options for Z.AI's mixed native and Anthropic transports. */
+export declare function zaiModelManagerOptions(_config?: ZaiModelManagerConfig): ModelManagerOptions<"anthropic-messages" | "openai-completions">;

package/dist/types/types.d.ts

@@ -29,6 +29,11 @@ export interface ThinkingConfig {
      * 5). Also implies native interleaved thinking — no beta header needed.
      */
     supportsDisplay?: boolean;
+    /**
+     * Thinking signatures bind each block to its preceding conversation prefix.
+     * Requests that rewrite the prefix must opt into the provider's drop/error control.
+     */
+    prefixBinding?: boolean;
     /**
      * Per-effort upstream wire-id routing for collapsed effort-tier variants
      * (`compat/collapse.ts`). Keyed by pi effort; `"off"` applies when
@@ -437,12 +442,20 @@ export interface AnthropicCompat {
     supportsLongCacheRetention?: boolean;
     /**
      * Whether mid-conversation `role: "system"` messages are accepted in the
-     * `messages` array (Claude Opus 4.8+ and Claude Fable/Mythos 5 on the
-     * first-party Claude API and Claude Platform on AWS). When unset,
-     * auto-detected from the model id and base URL. Not available on Bedrock,
-     * Vertex AI, or Microsoft Foundry.
+     * `messages` array. When unset, auto-detected from model and deployment policy.
      */
     supportsMidConversationSystem?: boolean;
+    /** Whether turn-scoped system messages accept `clear_at`. */
+    supportsTurnScopedSystem?: boolean;
+    /** Whether tool availability can change through system-message tool references. */
+    supportsMidConversationToolChanges?: boolean;
+    /** Whether effort can change through a per-message `output_config`. */
+    supportsPerMessageEffort?: boolean;
+    /**
+     * Whether the endpoint accepts `thinking.block_binding` and reports
+     * `input_transformations` under the thinking-binding-controls beta.
+     */
+    supportsThinkingBindingControls?: boolean;
     /**
      * Whether the model accepts a forced `tool_choice` (`{ type: "any" }` or
      * `{ type: "tool", name }`). Claude Fable/Mythos 5 reject forced tool use
@@ -866,6 +879,8 @@ export interface Model<TApi extends Api = Api> {
      * inferring it from the transport API.
      */
     requiresGlyphTokenization?: boolean;
+    /** Whether this model requires Cursor's tool-schema combiner projection. */
+    requiresCursorToolSchemaProjection?: boolean;
     /**
      * Model id to send on the wire when it differs from `id`. Used by catalog
      * variants that present one upstream model under several local entries —
@@ -1020,7 +1035,7 @@ export interface Model<TApi extends Api = Api> {
  * vocabulary of `buildModel`. Identical to `Model` except `compat` carries the
  * sparse override shape and nothing is resolved yet.
  */
-export interface ModelSpec<TApi extends Api = Api> extends Omit<Model<TApi>, "compat" | "identity" | "compatConfig" | "requiresGlyphTokenization" | "supportsComputerUseConfig"> {
+export interface ModelSpec<TApi extends Api = Api> extends Omit<Model<TApi>, "compat" | "identity" | "compatConfig" | "requiresGlyphTokenization" | "requiresCursorToolSchemaProjection" | "supportsComputerUseConfig"> {
     /** Sparse compatibility overrides; resolved into `Model.compat` by `buildModel`. */
     compat?: CompatConfigOf<TApi>;
 }

package/package.json

@@ -1,7 +1,7 @@
 {
 	"type": "module",
 	"name": "@oh-my-pi/pi-catalog",
-	"version": "18.1.1",
+	"version": "18.1.3",
 	"description": "Model catalog for omp: bundled model database, provider discovery descriptors, model identity, classification, and equivalence",
 	"homepage": "https://omp.sh",
 	"author": "Stencil Labs, Inc.",
@@ -35,12 +35,12 @@
 		"gen:proto": "bun scripts/generate-protocols.ts"
 	},
 	"dependencies": {
-		"@oh-my-pi/omptype": "18.1.1",
-		"@oh-my-pi/pi-utils": "18.1.1"
+		"@oh-my-pi/omptype": "18.1.3",
+		"@oh-my-pi/pi-utils": "18.1.3"
 	},
 	"devDependencies": {
 		"@bgotink/kdl": "0.4.0",
-		"@oh-my-pi/pi-ai": "18.1.1",
+		"@oh-my-pi/pi-ai": "18.1.3",
 		"@types/bun": "^1.3.14"
 	},
 	"engines": {

package/src/build.ts

@@ -32,7 +32,8 @@ function isInputModalities(value: unknown): value is ("text" | "image")[] {
  * Applies resolved catalog-data axes onto the model: reviewed metadata
  * corrections (`cost-patch`, `limits-patch`, `long-context-cost`,
  * `context-window-floor`) overwrite upstream values; selection metadata
- * (`priority`, `apply-patch-tool-type`, `service-tier-cost`) is rule-owned;
+ * (`priority`, `apply-patch-tool-type`, `service-tier-cost`,
+ * `requires-cursor-tool-schema-projection`) is rule-owned;
  * `context-promotion-target` fills only when the spec left it unset.
  */
 function applyCatalogAssignments<TApi extends Api>(model: Model<TApi>, catalog: Record<string, unknown>): void {
@@ -51,6 +52,12 @@ function applyCatalogAssignments<TApi extends Api>(model: Model<TApi>, catalog:
 	if (applyPatchToolType === "freeform" || applyPatchToolType === "function") {
 		model.applyPatchToolType = applyPatchToolType;
 	}
+	const requiresCursorToolSchemaProjection = catalog.requiresCursorToolSchemaProjection;
+	if (requiresCursorToolSchemaProjection === true) {
+		model.requiresCursorToolSchemaProjection = true;
+	} else {
+		delete model.requiresCursorToolSchemaProjection;
+	}
 	const contextPromotionTarget = catalog.contextPromotionTarget;
 	if (typeof contextPromotionTarget === "string" && model.contextPromotionTarget === undefined) {
 		model.contextPromotionTarget = contextPromotionTarget;

package/src/compat/axes.ts

@@ -193,6 +193,10 @@ export const AXES: Readonly<Record<string, AxisDef>> = {
 	"supports-eager-tool-input-streaming": wire("supportsEagerToolInputStreaming", ["anthropic"]),
 	"supports-long-cache-retention": wire("supportsLongCacheRetention", ["anthropic"]),
 	"supports-mid-conversation-system": wire("supportsMidConversationSystem", ["anthropic"]),
+	"supports-mid-conversation-tool-changes": wire("supportsMidConversationToolChanges", ["anthropic"]),
+	"supports-per-message-effort": wire("supportsPerMessageEffort", ["anthropic"]),
+	"supports-thinking-binding-controls": wire("supportsThinkingBindingControls", ["anthropic"]),
+	"supports-turn-scoped-system": wire("supportsTurnScopedSystem", ["anthropic"]),
 
 	// ── wire: bedrock-converse-stream ──
 	"prompt-cache-maximum-checkpoints": wire("promptCacheMaximumCheckpoints", ["bedrock"]),
@@ -239,6 +243,7 @@ export const AXES: Readonly<Record<string, AxisDef>> = {
 		values: THINKING_MODES,
 	},
 	"thinking-requires-effort": { key: "requiresEffort", set: "thinking", shape: "scalar" },
+	"thinking-prefix-binding": { key: "prefixBinding", set: "thinking", shape: "scalar" },
 	"thinking-suppress-when-off": { key: "suppressWhenOff", set: "thinking", shape: "scalar" },
 	"thinking-supports-display": { key: "supportsDisplay", set: "thinking", shape: "scalar" },
 
@@ -257,6 +262,11 @@ export const AXES: Readonly<Record<string, AxisDef>> = {
 	"limits-patch": { key: "limitsPatch", set: "catalog", shape: "object" },
 	"long-context-cost": { key: "longContext", set: "catalog", shape: "object" },
 	"long-usage-limit-fallback": { key: "longUsageLimitFallback", set: "catalog", shape: "scalar" },
+	"requires-cursor-tool-schema-projection": {
+		key: "requiresCursorToolSchemaProjection",
+		set: "catalog",
+		shape: "scalar",
+	},
 	priority: { key: "priority", set: "catalog", shape: "scalar" },
 	"service-tier-cost": { key: "serviceTierCost", set: "catalog", shape: "object" },
 };

package/src/compat/resolve.ts

@@ -846,7 +846,11 @@ function resolveAnthropicPolicy(
 		allowAnthropicHeaderOverrides: false,
 		supportsEagerToolInputStreaming: official,
 		supportsLongCacheRetention: official,
-		supportsMidConversationSystem: official && facts.anthropicAdaptiveGenAtLeast("4.8"),
+		supportsMidConversationSystem: official && !facts.family("sonnet") && facts.anthropicAdaptiveGenAtLeast("4.8"),
+		supportsTurnScopedSystem: false,
+		supportsMidConversationToolChanges: false,
+		supportsPerMessageEffort: false,
+		supportsThinkingBindingControls: false,
 		supportsForcedToolChoice: !requiresThinkingEnabled && !facts.family("fable", "mythos"),
 		supportsSamplingParams: !facts.anthropicAdaptiveGenAtLeast("4.7"),
 		requiresToolResultId: false,
@@ -1012,6 +1016,7 @@ interface RuleThinking {
 	requiresEffort?: boolean;
 	suppressWhenOff?: boolean;
 	supportsDisplay?: boolean;
+	prefixBinding?: boolean;
 }
 
 function readRuleThinking(axes: ResolvedAxes): RuleThinking {
@@ -1030,6 +1035,7 @@ function readRuleThinking(axes: ResolvedAxes): RuleThinking {
 	if (typeof raw.requiresEffort === "boolean") out.requiresEffort = raw.requiresEffort;
 	if (typeof raw.suppressWhenOff === "boolean") out.suppressWhenOff = raw.suppressWhenOff;
 	if (typeof raw.supportsDisplay === "boolean") out.supportsDisplay = raw.supportsDisplay;
+	if (typeof raw.prefixBinding === "boolean") out.prefixBinding = raw.prefixBinding;
 	return out;
 }
 
@@ -1085,6 +1091,7 @@ function resolveThinkingPolicy<TApi extends Api>(
 	if (rule.effortBudgets !== undefined) config.effortBudgets = rule.effortBudgets;
 	const supportsDisplay = rule.supportsDisplay ?? defaultSupportsDisplay(spec, facts);
 	if (supportsDisplay) config.supportsDisplay = true;
+	if (rule.prefixBinding) config.prefixBinding = true;
 	const requiresEffort =
 		rule.requiresEffort ?? (impliesMandatoryReasoning(facts, spec.id) || isQwenTemplateReasoningEffortCompat(compat));
 	if (requiresEffort) config.requiresEffort = true;
@@ -1142,7 +1149,8 @@ function fillExplicitThinking<TApi extends Api>(
 		(rule.requiresEffort ??
 			(impliesMandatoryReasoning(facts, spec.id) || isQwenTemplateReasoningEffortCompat(compat)));
 	const needsDefaultLevel = thinking.defaultLevel === undefined && rule.defaultLevel !== undefined;
-	if (effortMap === undefined && !needsDisplay && !needsRequiresEffort && !needsDefaultLevel) {
+	const needsPrefixBinding = thinking.prefixBinding === undefined && rule.prefixBinding === true;
+	if (effortMap === undefined && !needsDisplay && !needsRequiresEffort && !needsDefaultLevel && !needsPrefixBinding) {
 		return thinking;
 	}
 	const filled: ThinkingConfig = { ...thinking };
@@ -1150,6 +1158,7 @@ function fillExplicitThinking<TApi extends Api>(
 	if (needsDisplay) filled.supportsDisplay = true;
 	if (needsDefaultLevel && rule.defaultLevel !== undefined) filled.defaultLevel = rule.defaultLevel;
 	if (needsRequiresEffort) filled.requiresEffort = true;
+	if (needsPrefixBinding) filled.prefixBinding = true;
 	return filled;
 }
 

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

@@ -1,6 +1,49 @@
 // Model-lineage compat for "anthropic"; regenerated from the frozen census using class taxonomy selectors.
 
 class "anthropic" {
+	family "fable" {
+		revision ">=5.1" {
+			// Fable 5.1+ binds signed thinking to the exact preceding conversation.
+			thinking-prefix-binding #true
+		}
+	}
+	on "anthropic" "cloudflare-ai-gateway" "google-vertex" {
+		family "fable" {
+			revision ">=5" {
+				supports-mid-conversation-system #true
+				supports-mid-conversation-tool-changes #true
+				supports-turn-scoped-system #true
+			}
+			revision ">=5.1" {
+				supports-per-message-effort #true
+			}
+		}
+		family "mythos" {
+			revision ">=5" {
+				supports-mid-conversation-system #true
+				supports-mid-conversation-tool-changes #true
+				supports-turn-scoped-system #true
+			}
+			revision ">=5.1" {
+				supports-per-message-effort #true
+			}
+		}
+		family "fable" {
+			revision ">=5.1" {
+				supports-thinking-binding-controls #true
+			}
+		}
+		family "opus" {
+			revision ">=4.8" {
+				supports-mid-conversation-system #true
+				supports-mid-conversation-tool-changes #true
+				supports-turn-scoped-system #true
+			}
+			revision ">=5" {
+				supports-per-message-effort #true
+			}
+		}
+	}
 	on "aimlapi" "gitlab-duo" "kilo" "nanogpt" "venice" {
 		thinking-efforts "minimal" "low" "medium" "high" "xhigh"
 	}

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

@@ -66,17 +66,32 @@ class "gemini" {
 			}
 		}
 	}
+	// Gemini 3.7 Flash is the one Flash revision whose native thinkingLevel does
+	// not accept `minimal` (400 THINKING_LEVEL_MINIMAL). Only the direct
+	// google-level transports emit `thinkingLevel` on the wire, so the tier is
+	// dropped there; budget (anthropic-messages) and reasoning_effort resellers
+	// never send the rejected value and keep the four-tier ladder. The collapsed
+	// google-antigravity / google-gemini-cli variants remap minimal -> LOW and
+	// carry their own explicit ladder, so they are unaffected.
+	// https://ai.google.dev/gemini-api/docs/generate-content/thinking#thinking-levels
+	on "google" "google-vertex" "opencode-zen" {
+		family "flash" {
+			revision ">=3.7 <3.8" {
+				thinking-efforts "low" "medium" "high"
+			}
+		}
+	}
 
-	// Gemini 3 tool calls require the sentinel signature on unsigned replay.
 	revision ">=3" {
 		// Replaces the Google multimodal function-response baseline.
 		multimodal-function-response #true
-		requires-skip-thought-signature #true
 		// Replaces the Gemini 3+ mandatory-reasoning fallback.
 		thinking-requires-effort #true
 	}
+	// Only public-Gemini-compatible transports accept the bypass sentinel for unsigned calls.
 	on "google" "opencode-zen" {
 		revision ">=3" {
+			requires-skip-thought-signature #true
 			supports-function-part-id #true
 		}
 	}

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

@@ -10,7 +10,10 @@ class "glm" {
 	// GLM-5.2 exposes the default lower tiers plus a genuine max on generic
 	// OpenAI-compatible hosts; the zai-dialect and Anthropic-route hosts
 	// (Z.ai, Zhipu, Umans, Ollama Cloud, Baseten) serve only high/max.
-	revision ">=5.2 <5.3" {
+	// priority breaks the intentional overlap with provider-level blanket GLM
+	// ladders authored for the pre-5.2 census (e.g. alibaba-coding-plan.kdl),
+	// so the 5.2 lineage ladder wins on generic hosts that lack a 5.2 residue.
+	revision ">=5.2 <5.3" priority=10 {
 		thinking-efforts "minimal" "low" "medium" "high" "max"
 	}
 	on "zai" "zhipu-coding-plan" {

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

@@ -2,6 +2,11 @@
 
 provider "cursor" {
 	thinking-mode "effort"
+	class "anthropic" {
+		family "fable" {
+			requires-cursor-tool-schema-projection #true
+		}
+	}
 	// GetUsableModels advertises no input modalities or context windows.
 	// K3 (including the bare `k3` alias Cursor serves) is natively
 	// multimodal and 1M-context.

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

@@ -25,8 +25,12 @@ provider "zai" {
 	}
 	// GLM-5.3-Flash is the first mandatory-thinking flash SKU. Its exact
 	// deployment keeps the Anthropic budget+effort wire mode instead of the
-	// older optional-thinking flash family's budget-only mode.
+	// older optional-thinking flash family's budget-only mode. The SKU rides
+	// Z.AI's native OpenAI-completions endpoint, so clamp the sent max_tokens
+	// to the advertised 131K cap rather than the 64K OpenAI default (the
+	// axis is a no-op on the Anthropic route).
 	models "glm-5.3-flash" {
 		thinking-mode "anthropic-budget-effort"
+		clamp-output-to-model-max #true
 	}
 }

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

@@ -158,6 +158,11 @@ behavior {
 	api-routes provider="zenmux" default="openai-completions" {
 		route "anthropic-messages" prefix="anthropic/"
 	}
+	// GLM-5.3-Flash is available on Z.AI's native chat-completions API but
+	// not its Anthropic-compatible coding endpoint (#10539).
+	api-routes provider="zai" default="anthropic-messages" {
+		route "openai-completions" exact="glm-5.3-flash"
+	}
 	// OpenCode Zen: models.dev declares these with `@ai-sdk/anthropic`, but the
 	// gateway serves them only at /v1/chat/completions (#1617).
 	api-routes provider="opencode-zen" {