Enable tool streaming for newer Z.ai models (#2732)

packages/ai/scripts/generate-models.ts

@@ -55,6 +55,7 @@ const COPILOT_STATIC_HEADERS = {
 
 const AI_GATEWAY_MODELS_URL = "https://ai-gateway.vercel.sh/v1";
 const AI_GATEWAY_BASE_URL = "https://ai-gateway.vercel.sh";
+const ZAI_TOOL_STREAM_UNSUPPORTED_MODELS = new Set(["glm-4.5", "glm-4.5-air", "glm-4.5-flash", "glm-4.5v"]);
 
 async function fetchOpenRouterModels(): Promise<Model<any>[]> {
 	try {
@@ -379,28 +380,29 @@ async function loadModelsDevData(): Promise<Model<any>[]> {
 			for (const [modelId, model] of Object.entries(data.zai.models)) {
 				const m = model as ModelsDevModel;
 				if (m.tool_call !== true) continue;
-				const supportsImage = m.modalities?.input?.includes("image")
+				const supportsImage = m.modalities?.input?.includes("image");
 
 				models.push({
-				id: modelId,
-				name: m.name || modelId,
-				api: "openai-completions",
-				provider: "zai",
-				baseUrl: "https://api.z.ai/api/coding/paas/v4",
-				reasoning: m.reasoning === true,
-				input: supportsImage ? ["text", "image"] : ["text"],
-				cost: {
-					input: m.cost?.input || 0,
-					output: m.cost?.output || 0,
-					cacheRead: m.cost?.cache_read || 0,
-					cacheWrite: m.cost?.cache_write || 0,
-				},
-				compat: {
-					supportsDeveloperRole: false,
-					thinkingFormat: "zai",
-				},
-				contextWindow: m.limit?.context || 4096,
-				maxTokens: m.limit?.output || 4096,
+					id: modelId,
+					name: m.name || modelId,
+					api: "openai-completions",
+					provider: "zai",
+					baseUrl: "https://api.z.ai/api/coding/paas/v4",
+					reasoning: m.reasoning === true,
+					input: supportsImage ? ["text", "image"] : ["text"],
+					cost: {
+						input: m.cost?.input || 0,
+						output: m.cost?.output || 0,
+						cacheRead: m.cost?.cache_read || 0,
+						cacheWrite: m.cost?.cache_write || 0,
+					},
+					compat: {
+						supportsDeveloperRole: false,
+						thinkingFormat: "zai",
+						...(!ZAI_TOOL_STREAM_UNSUPPORTED_MODELS.has(modelId) ? { zaiToolStream: true } : {}),
+					},
+					contextWindow: m.limit?.context || 4096,
+					maxTokens: m.limit?.output || 4096,
 				});
 			}
 		}

packages/ai/src/providers/openai-completions.ts

@@ -395,6 +395,9 @@ function buildParams(model: Model<"openai-completions">, context: Context, optio
 
 	if (context.tools) {
 		params.tools = convertTools(context.tools, compat);
+		if (compat.zaiToolStream) {
+			(params as any).tool_stream = true;
+		}
 	} else if (hasToolHistory(context.messages)) {
 		// Anthropic (via LiteLLM/proxy) requires tools param when conversation has tool_calls/tool_results
 		params.tools = [];
@@ -835,6 +838,7 @@ function detectCompat(model: Model<"openai-completions">): Required<OpenAIComple
 				: "openai",
 		openRouterRouting: {},
 		vercelGatewayRouting: {},
+		zaiToolStream: false,
 		supportsStrictMode: true,
 	};
 }
@@ -861,6 +865,7 @@ function getCompat(model: Model<"openai-completions">): Required<OpenAICompletio
 		thinkingFormat: model.compat.thinkingFormat ?? detected.thinkingFormat,
 		openRouterRouting: model.compat.openRouterRouting ?? {},
 		vercelGatewayRouting: model.compat.vercelGatewayRouting ?? detected.vercelGatewayRouting,
+		zaiToolStream: model.compat.zaiToolStream ?? detected.zaiToolStream,
 		supportsStrictMode: model.compat.supportsStrictMode ?? detected.supportsStrictMode,
 	};
 }

packages/ai/src/types.ts

@@ -277,6 +277,8 @@ export interface OpenAICompletionsCompat {
 	openRouterRouting?: OpenRouterRouting;
 	/** Vercel AI Gateway routing preferences. Only used when baseUrl points to Vercel AI Gateway. */
 	vercelGatewayRouting?: VercelGatewayRouting;
+	/** Whether z.ai supports top-level `tool_stream: true` for streaming tool call deltas. Default: false. */
+	zaiToolStream?: boolean;
 	/** Whether the provider supports the `strict` field in tool definitions. Default: true. */
 	supportsStrictMode?: boolean;
 }

packages/ai/test/openai-completions-tool-choice.test.ts

@@ -200,6 +200,159 @@ describe("openai-completions tool_choice", () => {
 		expect(params.reasoning_effort).toBe("medium");
 	});
 
+	it("enables tool_stream for supported z.ai models with tools", async () => {
+		const model = getModel("zai", "glm-5")!;
+		const tools: Tool[] = [
+			{
+				name: "ping",
+				description: "Ping tool",
+				parameters: Type.Object({
+					ok: Type.Boolean(),
+				}),
+			},
+		];
+		let payload: unknown;
+
+		await streamSimple(
+			model,
+			{
+				messages: [
+					{
+						role: "user",
+						content: "Call ping with ok=true",
+						timestamp: Date.now(),
+					},
+				],
+				tools,
+			},
+			{
+				apiKey: "test",
+				onPayload: (params: unknown) => {
+					payload = params;
+				},
+			},
+		).result();
+
+		const params = (payload ?? mockState.lastParams) as { tool_stream?: boolean };
+		expect(params.tool_stream).toBe(true);
+	});
+
+	it("stores z.ai tool_stream support in model compat metadata", () => {
+		expect(getModel("zai", "glm-5")?.compat?.zaiToolStream).toBe(true);
+		expect(getModel("zai", "glm-4.7")?.compat?.zaiToolStream).toBe(true);
+		expect(getModel("zai", "glm-4.7-flash")?.compat?.zaiToolStream).toBe(true);
+		expect(getModel("zai", "glm-4.6v")?.compat?.zaiToolStream).toBe(true);
+		expect(getModel("zai", "glm-4.5-air")?.compat?.zaiToolStream).toBeUndefined();
+	});
+
+	it("omits tool_stream for unsupported z.ai models", async () => {
+		const model = getModel("zai", "glm-4.5-air")!;
+		const tools: Tool[] = [
+			{
+				name: "ping",
+				description: "Ping tool",
+				parameters: Type.Object({
+					ok: Type.Boolean(),
+				}),
+			},
+		];
+		let payload: unknown;
+
+		await streamSimple(
+			model,
+			{
+				messages: [
+					{
+						role: "user",
+						content: "Call ping with ok=true",
+						timestamp: Date.now(),
+					},
+				],
+				tools,
+			},
+			{
+				apiKey: "test",
+				onPayload: (params: unknown) => {
+					payload = params;
+				},
+			},
+		).result();
+
+		const params = (payload ?? mockState.lastParams) as { tool_stream?: boolean };
+		expect(params.tool_stream).toBeUndefined();
+	});
+
+	it("respects explicit z.ai tool_stream compat override", async () => {
+		const baseModel = getModel("zai", "glm-4.5-air")!;
+		const model = {
+			...baseModel,
+			compat: {
+				...baseModel.compat,
+				zaiToolStream: true,
+			},
+		} as const;
+		const tools: Tool[] = [
+			{
+				name: "ping",
+				description: "Ping tool",
+				parameters: Type.Object({
+					ok: Type.Boolean(),
+				}),
+			},
+		];
+		let payload: unknown;
+
+		await streamSimple(
+			model,
+			{
+				messages: [
+					{
+						role: "user",
+						content: "Call ping with ok=true",
+						timestamp: Date.now(),
+					},
+				],
+				tools,
+			},
+			{
+				apiKey: "test",
+				onPayload: (params: unknown) => {
+					payload = params;
+				},
+			},
+		).result();
+
+		const params = (payload ?? mockState.lastParams) as { tool_stream?: boolean };
+		expect(params.tool_stream).toBe(true);
+	});
+
+	it("omits tool_stream when no tools are provided", async () => {
+		const model = getModel("zai", "glm-5")!;
+		let payload: unknown;
+
+		await streamSimple(
+			model,
+			{
+				messages: [
+					{
+						role: "user",
+						content: "Hi",
+						timestamp: Date.now(),
+					},
+				],
+			},
+			{
+				apiKey: "test",
+				onPayload: (params: unknown) => {
+					payload = params;
+				},
+			},
+		).result();
+
+		const params = (payload ?? mockState.lastParams) as { tool_stream?: boolean };
+		expect(params.tool_stream).toBeUndefined();
+	});
+
 	it("maps non-standard provider finish_reason values to stopReason error", async () => {
 		mockState.chunks = [
 			{

packages/ai/test/openai-completions-tool-result-images.test.ts

@@ -32,6 +32,7 @@ const compat: Required<OpenAICompletionsCompat> = {
 	thinkingFormat: "openai",
 	openRouterRouting: {},
 	vercelGatewayRouting: {},
+	zaiToolStream: false,
 	supportsStrictMode: true,
 };