Replace GLM 5.2 with GLM 5.3 Flash

packages/workshop-backend/__tests__/ai-models.test.ts

@@ -526,6 +526,18 @@ describe("PDF attachment bridging", () => {
     return JSON.parse(capturedRequests[0].body);
   }
 
+  it("preserves image input for GLM 5.3 Flash", async () => {
+    const handle = getModel(env(),
+        {...WORKERS_AI_CONFIG, model: "@cf/zai-org/glm-5.3-flash"}, INITIATOR);
+    expect(handle.model).toMatchObject({reasoning: true, input: ["text", "image"]});
+    const body = await capturePdfRequest(handle) as {
+      messages: { content: { image_url?: { url: string } }[] }[],
+    };
+    expect(body.messages[0].content).toContainEqual(expect.objectContaining({
+      image_url: { url: "data:image/png;base64,iVBOR" },
+    }));
+  }, 15000);
+
   it("sends Anthropic PDFs as document blocks", async () => {
     const handle = getModel(env(), ANTHROPIC_CONFIG, INITIATOR);
     const body = await capturePdfRequest(handle) as

packages/workshop-backend/src/ai-models.ts

@@ -125,14 +125,25 @@ const API_STREAMS: Record<string, StreamFunction<Api, SimpleStreamOptions>> = {
 
 const ZERO_COST: ModelCost = { input: 0, output: 0, cacheRead: 0, cacheWrite: 0 };
 
+// Remove when pi's bundled Workers AI catalog includes this model.
+const GLM_5_3_FLASH_FALLBACK: Model<"openai-completions"> = {
+  id: "@cf/zai-org/glm-5.3-flash", name: "GLM 5.3 Flash",
+  api: "openai-completions", provider: "cloudflare-workers-ai",
+  baseUrl: "https://api.cloudflare.com/client/v4/accounts/{CLOUDFLARE_ACCOUNT_ID}/ai/v1",
+  reasoning: true, input: ["text", "image"],
+  cost: {input: 0.15, output: 0.5, cacheRead: 0.03, cacheWrite: 0},
+  contextWindow: 1048576, maxTokens: WORKERS_AI_OUTPUT_LIMIT,
+};
+
 // Consult pi's builtin catalog for cost/compat metadata of a known model id. Unknown models are
 // fine (synthesized with zero cost). Import per-provider, not providers/all.
 function catalogModel(provider: AiModelConfig["provider"], modelId: string): Model<Api> | undefined {
   switch (provider) {
     case "anthropic": return (ANTHROPIC_MODELS as Record<string, Model<Api>>)[modelId];
     case "openai": return (OPENAI_MODELS as Record<string, Model<Api>>)[modelId];
     case "google": return (GOOGLE_MODELS as Record<string, Model<Api>>)[modelId];
-    case "cloudflare": return (CLOUDFLARE_WORKERS_AI_MODELS as Record<string, Model<Api>>)[modelId];
+    case "cloudflare": return (CLOUDFLARE_WORKERS_AI_MODELS as Record<string, Model<Api>>)[modelId] ??
+        (modelId === GLM_5_3_FLASH_FALLBACK.id ? GLM_5_3_FLASH_FALLBACK : undefined);
     case "ollama": return undefined;
     default: return undefined;
   }

packages/workshop-shared/src/api.ts

@@ -1192,8 +1192,9 @@ export const SUGGESTED_MODELS: Record<
       name: "Kimi K2.7 Code (Workers AI)", contextWindow: 262144,
       outputLimit: WORKERS_AI_OUTPUT_LIMIT,
     },
-    "@cf/zai-org/glm-5.2": {
-      name: "GLM 5.2 (Workers AI)", contextWindow: 262144, outputLimit: WORKERS_AI_OUTPUT_LIMIT,
+    "@cf/zai-org/glm-5.3-flash": {
+      name: "GLM 5.3 Flash (Workers AI)", contextWindow: 1048576,
+      outputLimit: WORKERS_AI_OUTPUT_LIMIT,
     },
     "@cf/deepseek-ai/deepseek-v4-pro-0813": {
       name: "DeepSeek V4 Pro 0813 (Workers AI)", contextWindow: 1048576,