feat(catalog): add GLM-5.3 support with uniform low/high/max effort ladder and mandatory thinking

GLM-5.3 introduces three key API changes from GLM-5.2:
- Uniform wire-exact low/high/max reasoning_effort ladder on every host
  (replacing GLM-5.2's host-specific dialects)
- Thinking can no longer be disabled (thinking.type must always be "enabled")
- Default effort is max

Changes:
- Add isGlm53ReasoningEffortModelId classifier (>=5.3, base/air/turbo, non-vision)
- getModelDefinedEfforts: GLM-5.3 returns LOW_HIGH_MAX uniformly
- impliesMandatoryReasoning: GLM-5.3 floors thinking-off to lowest effort
- deriveThinking/fillThinkingWireDefaults: defaultLevel=max for GLM-5.3
- generated-policies: pin glm-5.3 to 1M context (zai + zhipu-coding-plan)
- descriptors: zai defaultModel -> glm-5.3
- generate-models: curated seed (glm-5.3 is live but not in /models discovery)
- models.json: bundled glm-5.3 entry
- Tests: catalog thinking-metadata + AI wire-mapping (5 new tests)

packages/catalog/scripts/generate-models.ts

@@ -555,6 +555,24 @@ async function generateModels() {
 	// Mythos 5). Deduped behind upstream entries; metadata is pinned in
 	// applyAnthropicCatalogPolicy.
 	allModels.push(...ANTHROPIC_CURATED_FALLBACK_MODELS);
+	// Seed GLM-5.3 on the z.AI provider. GLM-5.3 is live on the Anthropic and
+	// coding endpoints but not yet advertised in `/v1/models` (which still tops
+	// out at glm-5.2), so endpoint discovery misses it. The zai provider is not
+	// authoritative, so the seed survives regeneration; thinking metadata
+	// (low/high/max uniform ladder, mandatory reasoning, defaultLevel=max) is
+	// derived by rebakeModelThinking from the identity classifiers.
+	allModels.push({
+		id: "glm-5.3",
+		name: "GLM-5.3",
+		api: "anthropic-messages",
+		provider: "zai",
+		baseUrl: "https://api.z.ai/api/anthropic",
+		reasoning: true,
+		input: ["text"],
+		cost: { input: 1.4, output: 4.4, cacheRead: 0.26, cacheWrite: 0 },
+		contextWindow: 1_000_000,
+		maxTokens: 131_072,
+	} as ModelSpec<"anthropic-messages">);
 	// Seed Meta's documented Muse model so first-run selection does not depend on
 	// credentials or live discovery.
 	allModels.push(...META_MUSE_STATIC_MODELS);

packages/catalog/scripts/generated-policies.ts

@@ -367,9 +367,13 @@ function applyGeneratedModelPolicy(model: ModelSpec<Api>): void {
 		model.omitMaxOutputTokens = true;
 	}
 
-	// GLM Coding Plan: GLM-5.2 is the selectable 1M served id; pin it so
+	// GLM Coding Plan: the selectable 1M-context served ids; pin them so
 	// endpoint discovery or older bundled fallbacks cannot regress to 200k.
-	if ((model.provider === "zai" || model.provider === "zhipu-coding-plan") && model.id === "glm-5.2") {
+	// GLM-5.3 succeeds GLM-5.2 with the same 1M context window.
+	if (
+		(model.provider === "zai" || model.provider === "zhipu-coding-plan") &&
+		(model.id === "glm-5.2" || model.id === "glm-5.3")
+	) {
 		model.contextWindow = 1_000_000;
 		model.maxTokens = 131_072;
 	}