feat: add GLM-5 model support to Z.ai provider (#11440)

packages/types/src/providers/zai.ts

@@ -120,6 +120,21 @@ export const internationalZAiModels = {
 		description:
 			"GLM-4.7 is Zhipu's latest model with built-in thinking capabilities enabled by default. It provides enhanced reasoning for complex tasks while maintaining fast response times.",
 	},
+	"glm-5": {
+		maxTokens: 16_384,
+		contextWindow: 202_752,
+		supportsImages: false,
+		supportsPromptCache: true,
+		supportsReasoningEffort: ["disable", "medium"],
+		reasoningEffort: "medium",
+		preserveReasoning: true,
+		inputPrice: 0.6,
+		outputPrice: 2.2,
+		cacheWritesPrice: 0,
+		cacheReadsPrice: 0.11,
+		description:
+			"GLM-5 is Zhipu's next-generation model with a 202k context window and built-in thinking capabilities. It delivers state-of-the-art reasoning, coding, and agentic performance.",
+	},
 	"glm-4.7-flash": {
 		maxTokens: 16_384,
 		contextWindow: 200_000,
@@ -281,6 +296,21 @@ export const mainlandZAiModels = {
 		description:
 			"GLM-4.7 is Zhipu's latest model with built-in thinking capabilities enabled by default. It provides enhanced reasoning for complex tasks while maintaining fast response times.",
 	},
+	"glm-5": {
+		maxTokens: 16_384,
+		contextWindow: 202_752,
+		supportsImages: false,
+		supportsPromptCache: true,
+		supportsReasoningEffort: ["disable", "medium"],
+		reasoningEffort: "medium",
+		preserveReasoning: true,
+		inputPrice: 0.29,
+		outputPrice: 1.14,
+		cacheWritesPrice: 0,
+		cacheReadsPrice: 0.057,
+		description:
+			"GLM-5 is Zhipu's next-generation model with a 202k context window and built-in thinking capabilities. It delivers state-of-the-art reasoning, coding, and agentic performance.",
+	},
 	"glm-4.7-flash": {
 		maxTokens: 16_384,
 		contextWindow: 204_800,

src/api/providers/__tests__/zai.spec.ts

@@ -121,6 +121,22 @@ describe("ZAiHandler", () => {
 			expect(model.info.preserveReasoning).toBe(true)
 		})
 
+		it("should return GLM-5 international model with thinking support", () => {
+			const testModelId: InternationalZAiModelId = "glm-5"
+			const handlerWithModel = new ZAiHandler({
+				apiModelId: testModelId,
+				zaiApiKey: "test-zai-api-key",
+				zaiApiLine: "international_coding",
+			})
+			const model = handlerWithModel.getModel()
+			expect(model.id).toBe(testModelId)
+			expect(model.info).toEqual(internationalZAiModels[testModelId])
+			expect(model.info.contextWindow).toBe(202_752)
+			expect(model.info.supportsReasoningEffort).toEqual(["disable", "medium"])
+			expect(model.info.reasoningEffort).toBe("medium")
+			expect(model.info.preserveReasoning).toBe(true)
+		})
+
 		it("should return GLM-4.5v international model with vision support", () => {
 			const testModelId: InternationalZAiModelId = "glm-4.5v"
 			const handlerWithModel = new ZAiHandler({
@@ -203,6 +219,22 @@ describe("ZAiHandler", () => {
 			expect(model.info.reasoningEffort).toBe("medium")
 			expect(model.info.preserveReasoning).toBe(true)
 		})
+
+		it("should return GLM-5 China model with thinking support", () => {
+			const testModelId: MainlandZAiModelId = "glm-5"
+			const handlerWithModel = new ZAiHandler({
+				apiModelId: testModelId,
+				zaiApiKey: "test-zai-api-key",
+				zaiApiLine: "china_coding",
+			})
+			const model = handlerWithModel.getModel()
+			expect(model.id).toBe(testModelId)
+			expect(model.info).toEqual(mainlandZAiModels[testModelId])
+			expect(model.info.contextWindow).toBe(202_752)
+			expect(model.info.supportsReasoningEffort).toEqual(["disable", "medium"])
+			expect(model.info.reasoningEffort).toBe("medium")
+			expect(model.info.preserveReasoning).toBe(true)
+		})
 	})
 
 	describe("International API", () => {
@@ -508,6 +540,74 @@ describe("ZAiHandler", () => {
 		})
 	})
 
+	describe("GLM-5 Thinking Mode", () => {
+		it("should enable thinking by default for GLM-5 (default reasoningEffort is medium)", async () => {
+			const handlerWithModel = new ZAiHandler({
+				apiModelId: "glm-5",
+				zaiApiKey: "test-zai-api-key",
+				zaiApiLine: "international_coding",
+			})
+
+			async function* mockFullStream() {
+				yield { type: "text-delta", text: "response" }
+			}
+
+			mockStreamText.mockReturnValue({
+				fullStream: mockFullStream(),
+				usage: Promise.resolve({ inputTokens: 0, outputTokens: 0 }),
+			})
+
+			const stream = handlerWithModel.createMessage("system prompt", [])
+			for await (const _chunk of stream) {
+				// drain
+			}
+
+			expect(mockStreamText).toHaveBeenCalledWith(
+				expect.objectContaining({
+					providerOptions: {
+						zhipu: {
+							thinking: { type: "enabled" },
+						},
+					},
+				}),
+			)
+		})
+
+		it("should disable thinking for GLM-5 when reasoningEffort is set to disable", async () => {
+			const handlerWithModel = new ZAiHandler({
+				apiModelId: "glm-5",
+				zaiApiKey: "test-zai-api-key",
+				zaiApiLine: "international_coding",
+				enableReasoningEffort: true,
+				reasoningEffort: "disable",
+			})
+
+			async function* mockFullStream() {
+				yield { type: "text-delta", text: "response" }
+			}
+
+			mockStreamText.mockReturnValue({
+				fullStream: mockFullStream(),
+				usage: Promise.resolve({ inputTokens: 0, outputTokens: 0 }),
+			})
+
+			const stream = handlerWithModel.createMessage("system prompt", [])
+			for await (const _chunk of stream) {
+				// drain
+			}
+
+			expect(mockStreamText).toHaveBeenCalledWith(
+				expect.objectContaining({
+					providerOptions: {
+						zhipu: {
+							thinking: { type: "disabled" },
+						},
+					},
+				}),
+			)
+		})
+	})
+
 	describe("completePrompt", () => {
 		it("should complete a prompt using generateText", async () => {
 			mockGenerateText.mockResolvedValue({

src/api/providers/zai.ts

@@ -115,8 +115,8 @@ export class ZAiHandler extends BaseProvider implements SingleCompletionHandler
 			toolChoice: mapToolChoice(metadata?.tool_choice),
 		}
 
-		// GLM-4.7 thinking mode: pass thinking parameter via providerOptions
-		const isThinkingModel = modelId === "glm-4.7" && Array.isArray(info.supportsReasoningEffort)
+		// Thinking mode: pass thinking parameter via providerOptions for models that support it (e.g. GLM-4.7, GLM-5)
+		const isThinkingModel = Array.isArray(info.supportsReasoningEffort)
 
 		if (isThinkingModel) {
 			const useReasoning = shouldUseReasoningEffort({ model: info, settings: this.options })