From bd28eb9f5bc614d0ca170cdcb2df556c6b6c945c Mon Sep 17 00:00:00 2001
From: Vincent Koc <vincentkoc@ieee.org>
Date: Sun, 22 Mar 2026 17:51:18 -0700
Subject: [PATCH] fix(zai): align remaining pi metadata

---
 extensions/zai/model-definitions.test.ts | 25 ++++++++++++++++++++++--
 extensions/zai/model-definitions.ts      | 18 ++++++++---------
 2 files changed, 32 insertions(+), 11 deletions(-)

diff --git a/extensions/zai/model-definitions.test.ts b/extensions/zai/model-definitions.test.ts
index 22e2bc7e484..615ccf3c91b 100644
--- a/extensions/zai/model-definitions.test.ts
+++ b/extensions/zai/model-definitions.test.ts
@@ -7,8 +7,8 @@ describe("zai model definitions", () => {
       id: "glm-5",
       reasoning: true,
       input: ["text"],
-      contextWindow: 204800,
-      maxTokens: 131072,
+      contextWindow: 202800,
+      maxTokens: 131100,
       cost: ZAI_DEFAULT_COST,
     });
   });
@@ -29,4 +29,25 @@ describe("zai model definitions", () => {
       cost: { input: 0.2, output: 1.1, cacheRead: 0.03, cacheWrite: 0 },
     });
   });
+
+  it("keeps the remaining GLM 4.7/5 pricing and token limits aligned with Pi", () => {
+    expect(buildZaiModelDefinition({ id: "glm-4.7-flash" })).toMatchObject({
+      id: "glm-4.7-flash",
+      cost: { input: 0.07, output: 0.4, cacheRead: 0, cacheWrite: 0 },
+      contextWindow: 200000,
+      maxTokens: 131072,
+    });
+    expect(buildZaiModelDefinition({ id: "glm-4.7-flashx" })).toMatchObject({
+      id: "glm-4.7-flashx",
+      cost: { input: 0.06, output: 0.4, cacheRead: 0.01, cacheWrite: 0 },
+      contextWindow: 200000,
+      maxTokens: 128000,
+    });
+    expect(buildZaiModelDefinition({ id: "glm-5-turbo" })).toMatchObject({
+      id: "glm-5-turbo",
+      contextWindow: 202800,
+      maxTokens: 131100,
+      cost: { input: 1.2, output: 4, cacheRead: 0.24, cacheWrite: 0 },
+    });
+  });
 });
diff --git a/extensions/zai/model-definitions.ts b/extensions/zai/model-definitions.ts
index a75fe2e0bc6..ce59be40f75 100644
--- a/extensions/zai/model-definitions.ts
+++ b/extensions/zai/model-definitions.ts
@@ -28,16 +28,16 @@ const ZAI_MODEL_CATALOG = {
     name: "GLM-5",
     reasoning: true,
     input: ["text"],
-    contextWindow: 204800,
-    maxTokens: 131072,
+    contextWindow: 202800,
+    maxTokens: 131100,
     cost: ZAI_DEFAULT_COST,
   },
   "glm-5-turbo": {
     name: "GLM-5 Turbo",
     reasoning: true,
     input: ["text"],
-    contextWindow: 200000,
-    maxTokens: 131072,
+    contextWindow: 202800,
+    maxTokens: 131100,
     cost: { input: 1.2, output: 4, cacheRead: 0.24, cacheWrite: 0 },
   },
   "glm-4.7": {
@@ -54,15 +54,15 @@ const ZAI_MODEL_CATALOG = {
     input: ["text"],
     contextWindow: 200000,
     maxTokens: 131072,
-    cost: { input: 0, output: 0, cacheRead: 0, cacheWrite: 0 },
+    cost: { input: 0.07, output: 0.4, cacheRead: 0, cacheWrite: 0 },
   },
   "glm-4.7-flashx": {
     name: "GLM-4.7 FlashX",
     reasoning: true,
     input: ["text"],
     contextWindow: 200000,
-    maxTokens: 131072,
-    cost: { input: 0, output: 0, cacheRead: 0, cacheWrite: 0 },
+    maxTokens: 128000,
+    cost: { input: 0.06, output: 0.4, cacheRead: 0.01, cacheWrite: 0 },
   },
   "glm-4.6": {
     name: "GLM-4.6",
@@ -148,7 +148,7 @@ export function buildZaiModelDefinition(params: {
     input:
       params.input ?? (catalog?.input ? ([...catalog.input] as ("text" | "image")[]) : ["text"]),
     cost: params.cost ?? catalog?.cost ?? ZAI_DEFAULT_COST,
-    contextWindow: params.contextWindow ?? catalog?.contextWindow ?? 204800,
-    maxTokens: params.maxTokens ?? catalog?.maxTokens ?? 131072,
+    contextWindow: params.contextWindow ?? catalog?.contextWindow ?? 202800,
+    maxTokens: params.maxTokens ?? catalog?.maxTokens ?? 131100,
   };
 }