@prestyj/core 5.12.0 → 5.13.0

package/dist/index.cjs

@@ -2304,11 +2304,16 @@ var MODELS = [
     maxThinkingLevel: "high"
   },
   // ── Z.AI (GLM) ─────────────────────────────────────────
-  // GLM-5.2: coding-first flagship with a usable 1M-token context window
-  // (5x jump over GLM-5.1's ~200K) and 131K max output. Released 2026-06-13.
+  // GLM-5.3 is the only GLM entry: it supersedes 5.2 (same GLM-5 base, all
+  // gains from post-training) and the coding endpoint already answers
+  // `glm-5.2` requests as glm-5.3, so the older ids were menu clutter that
+  // routed to strictly worse coding for the same plan quota.
+  // Released 2026-08-14; live on the coding endpoint (verified), while the
+  // standard paas API is still "coming soon". `max` is both the ceiling and
+  // Z.AI's own default — the rungs below it live in thinking-level.ts.
   {
-    id: "glm-5.2",
-    name: "GLM-5.2",
+    id: "glm-5.3",
+    name: "GLM-5.3",
     provider: "glm",
     contextWindow: 1e6,
     maxOutputTokens: 131072,
@@ -2316,43 +2321,7 @@ var MODELS = [
     supportsImages: false,
     supportsVideo: false,
     costTier: "medium",
-    maxThinkingLevel: "high"
-  },
-  {
-    id: "glm-5.1",
-    name: "GLM-5.1",
-    provider: "glm",
-    contextWindow: 204800,
-    maxOutputTokens: 131072,
-    supportsThinking: true,
-    supportsImages: false,
-    supportsVideo: false,
-    costTier: "medium",
-    maxThinkingLevel: "high"
-  },
-  {
-    id: "glm-4.7",
-    name: "GLM-4.7",
-    provider: "glm",
-    contextWindow: 2e5,
-    maxOutputTokens: 131072,
-    supportsThinking: true,
-    supportsImages: false,
-    supportsVideo: false,
-    costTier: "low",
-    maxThinkingLevel: "high"
-  },
-  {
-    id: "glm-4.7-flash",
-    name: "GLM-4.7 Flash",
-    provider: "glm",
-    contextWindow: 2e5,
-    maxOutputTokens: 131072,
-    supportsThinking: true,
-    supportsImages: false,
-    supportsVideo: false,
-    costTier: "low",
-    maxThinkingLevel: "high"
+    maxThinkingLevel: "max"
   },
   // ── MiniMax ────────────────────────────────────────────
   {
@@ -2497,7 +2466,7 @@ function getDefaultModel(provider) {
   if (provider === "xiaomi") return MODELS.find((m) => m.id === "mimo-v2.5-pro");
   if (provider === "openai") return MODELS.find((m) => m.id === "gpt-5.6-sol");
   if (provider === "gemini") return MODELS.find((m) => m.id === "gemini-3.1-flash-lite");
-  if (provider === "glm") return MODELS.find((m) => m.id === "glm-5.2");
+  if (provider === "glm") return MODELS.find((m) => m.id === "glm-5.3");
   if (provider === "moonshot") return MODELS.find((m) => m.id === "kimi-k3");
   if (provider === "minimax") return MODELS.find((m) => m.id === "MiniMax-M3");
   if (provider === "deepseek") return MODELS.find((m) => m.id === "deepseek-v4-pro");
@@ -2584,6 +2553,7 @@ var ANTHROPIC_ADAPTIVE_THINKING_LEVELS = [
   "max"
 ];
 var MOONSHOT_K3_THINKING_LEVELS = ["low", "high", "max"];
+var GLM_THINKING_LEVELS = ["low", "medium", "high", "xhigh", "max"];
 var LOCAL_THINKING_LEVELS = ["low", "medium", "high", "max"];
 function isOpenAIGptModel(provider, model) {
   return provider === "openai" && model.startsWith("gpt-");
@@ -2597,6 +2567,9 @@ function isXaiModel(provider) {
 function isMoonshotK3Model(provider, model) {
   return provider === "moonshot" && model === "kimi-k3";
 }
+function isGlmModel(provider) {
+  return provider === "glm";
+}
 function isAnthropicXhighModel(provider, model) {
   return provider === "anthropic" && /opus-5|opus-4-8|opus-4-7/.test(model);
 }
@@ -2628,6 +2601,11 @@ function getSupportedThinkingLevels(provider, model) {
     return XAI_THINKING_LEVELS.slice(0, maxIndex2 + 1);
   }
   if (isMoonshotK3Model(provider, model)) return MOONSHOT_K3_THINKING_LEVELS;
+  if (isGlmModel(provider)) {
+    const maxIndex2 = GLM_THINKING_LEVELS.indexOf(maxLevel);
+    if (maxIndex2 === -1) return GLM_THINKING_LEVELS;
+    return GLM_THINKING_LEVELS.slice(0, maxIndex2 + 1);
+  }
   if (!isOpenAIGptModel(provider, model)) return [maxLevel];
   const levels = model.startsWith("gpt-5.6-") ? OPENAI_GPT_56_THINKING_LEVELS : OPENAI_GPT_THINKING_LEVELS;
   const maxIndex = levels.indexOf(maxLevel);
@@ -2639,7 +2617,7 @@ function isThinkingLevelSupported(provider, model, level) {
 }
 function getNextThinkingLevel(provider, model, current) {
   const supportedLevels = getSupportedThinkingLevels(provider, model);
-  const shouldCycleLevels = isOpenAIGptModel(provider, model) || isAnthropicAdaptiveModel(provider, model) || isSakanaModel(provider) || isXaiModel(provider) || isMoonshotK3Model(provider, model) || // Local servers take a real effort level, not just on/off: Ollama accepts
+  const shouldCycleLevels = isOpenAIGptModel(provider, model) || isAnthropicAdaptiveModel(provider, model) || isSakanaModel(provider) || isXaiModel(provider) || isMoonshotK3Model(provider, model) || isGlmModel(provider) || // Local servers take a real effort level, not just on/off: Ollama accepts
   // low/medium/high on `reasoning_effort` (verified against 0.32) and the
   // other OpenAI-compatible servers use the same three. A model that can't
   // reason at all already has no supported levels, so it never gets here.