Add GLM 5

.changeset/zai-glm5-default-lines.md

@@ -0,0 +1,5 @@
+---
+"kilo-code": patch
+---
+
+Add support for GLM 5 and set Z.ai default to `glm-5` and align Z.ai API line model selection in VS Code and webview settings

packages/types/src/providers/zai.ts

@@ -6,11 +6,12 @@ import { ZaiApiLine } from "../provider-settings.js"
 // https://docs.z.ai/guides/llm/glm-4.5
 // https://docs.z.ai/guides/llm/glm-4.6
 // https://docs.z.ai/guides/llm/glm-4.7
+// https://docs.z.ai/guides/llm/glm-5 // kilocode_change
 // https://docs.z.ai/guides/overview/pricing
 // https://bigmodel.cn/pricing
 
 export type InternationalZAiModelId = keyof typeof internationalZAiModels
-export const internationalZAiDefaultModelId: InternationalZAiModelId = "glm-4.7"
+export const internationalZAiDefaultModelId: InternationalZAiModelId = "glm-5" // kilocode_change
 export const internationalZAiModels = {
 	"glm-4.5": {
 		maxTokens: 16_384,
@@ -157,6 +158,24 @@ export const internationalZAiModels = {
 		preferredIndex: 1,
 	},
 	// kilocode_change start
+	"glm-5": {
+		maxTokens: 131_072,
+		contextWindow: 200_000,
+		supportsImages: false,
+		supportsPromptCache: true,
+		supportsNativeTools: true,
+		defaultToolProtocol: "native",
+		supportsReasoningEffort: ["disable", "medium"],
+		reasoningEffort: "medium",
+		preserveReasoning: true,
+		inputPrice: 1,
+		outputPrice: 3.2,
+		cacheWritesPrice: 0,
+		cacheReadsPrice: 0.2,
+		description:
+			"GLM-5 is Z.AI's flagship text model with 200K context, 128K max output, thinking mode, function calling, and context caching.",
+		preferredIndex: 0,
+	},
 	"glm-4.7-flash": {
 		maxTokens: 16_384,
 		contextWindow: 200_000,
@@ -189,7 +208,7 @@ export const internationalZAiModels = {
 } as const satisfies Record<string, ModelInfo>
 
 export type MainlandZAiModelId = keyof typeof mainlandZAiModels
-export const mainlandZAiDefaultModelId: MainlandZAiModelId = "glm-4.7"
+export const mainlandZAiDefaultModelId: MainlandZAiModelId = "glm-5" // kilocode_change
 export const mainlandZAiModels = {
 	"glm-4.5": {
 		maxTokens: 16_384,
@@ -306,6 +325,24 @@ export const mainlandZAiModels = {
 		preferredIndex: 1,
 	},
 	// kilocode_change start
+	"glm-5": {
+		maxTokens: 131_072,
+		contextWindow: 200_000,
+		supportsImages: false,
+		supportsPromptCache: true,
+		supportsNativeTools: true,
+		defaultToolProtocol: "native",
+		supportsReasoningEffort: ["disable", "medium"],
+		reasoningEffort: "medium",
+		preserveReasoning: true,
+		inputPrice: 0.57,
+		outputPrice: 2.57,
+		cacheWritesPrice: 0,
+		cacheReadsPrice: 0.14,
+		description:
+			"GLM-5 is Z.AI's flagship text model with 200K context, 128K max output, thinking mode, function calling, and context caching.",
+		preferredIndex: 0,
+	},
 	"glm-4.7-flash": {
 		maxTokens: 16_384,
 		contextWindow: 204_800,

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

@@ -108,6 +108,25 @@ describe("ZAiHandler", () => {
 			expect(model.info.preserveReasoning).toBe(true)
 		})
 
+		// kilocode_change start
+		it("should return GLM-5 international model with documented limits", () => {
+			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(200_000)
+			expect(model.info.maxTokens).toBe(131_072)
+			expect(model.info.supportsReasoningEffort).toEqual(["disable", "medium"])
+			expect(model.info.reasoningEffort).toBe("medium")
+			expect(model.info.preserveReasoning).toBe(true)
+		})
+		// kilocode_change end
+
 		it("should return GLM-4.5v international model with vision support", () => {
 			const testModelId: InternationalZAiModelId = "glm-4.5v"
 			const handlerWithModel = new ZAiHandler({
@@ -203,6 +222,25 @@ describe("ZAiHandler", () => {
 			expect(model.info.reasoningEffort).toBe("medium")
 			expect(model.info.preserveReasoning).toBe(true)
 		})
+
+		// kilocode_change start
+		it("should return GLM-5 China model with documented limits", () => {
+			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(200_000)
+			expect(model.info.maxTokens).toBe(131_072)
+			expect(model.info.supportsReasoningEffort).toEqual(["disable", "medium"])
+			expect(model.info.reasoningEffort).toBe("medium")
+			expect(model.info.preserveReasoning).toBe(true)
+		})
+		// kilocode_change end
 	})
 
 	describe("International API", () => {
@@ -242,6 +280,23 @@ describe("ZAiHandler", () => {
 			expect(model.id).toBe(testModelId)
 			expect(model.info).toEqual(internationalZAiModels[testModelId])
 		})
+
+		// kilocode_change start
+		it("should return GLM-5 international API model with documented limits", () => {
+			const testModelId: InternationalZAiModelId = "glm-5"
+			const handlerWithModel = new ZAiHandler({
+				apiModelId: testModelId,
+				zaiApiKey: "test-zai-api-key",
+				zaiApiLine: "international_api",
+			})
+			const model = handlerWithModel.getModel()
+			expect(model.id).toBe(testModelId)
+			expect(model.info).toEqual(internationalZAiModels[testModelId])
+			expect(model.info.contextWindow).toBe(200_000)
+			expect(model.info.maxTokens).toBe(131_072)
+			expect(model.info.supportsReasoningEffort).toEqual(["disable", "medium"])
+		})
+		// kilocode_change end
 	})
 
 	describe("China API", () => {
@@ -281,6 +336,23 @@ describe("ZAiHandler", () => {
 			expect(model.id).toBe(testModelId)
 			expect(model.info).toEqual(mainlandZAiModels[testModelId])
 		})
+
+		// kilocode_change start
+		it("should return GLM-5 China API model with documented limits", () => {
+			const testModelId: MainlandZAiModelId = "glm-5"
+			const handlerWithModel = new ZAiHandler({
+				apiModelId: testModelId,
+				zaiApiKey: "test-zai-api-key",
+				zaiApiLine: "china_api",
+			})
+			const model = handlerWithModel.getModel()
+			expect(model.id).toBe(testModelId)
+			expect(model.info).toEqual(mainlandZAiModels[testModelId])
+			expect(model.info.contextWindow).toBe(200_000)
+			expect(model.info.maxTokens).toBe(131_072)
+			expect(model.info.supportsReasoningEffort).toEqual(["disable", "medium"])
+		})
+		// kilocode_change end
 	})
 
 	describe("Default behavior", () => {
@@ -414,7 +486,8 @@ describe("ZAiHandler", () => {
 		})
 	})
 
-	describe("GLM-4.7 Thinking Mode", () => {
+	// kilocode_change start
+	describe("Z.ai Thinking Mode", () => {
 		it("should enable thinking by default for GLM-4.7 (default reasoningEffort is medium)", async () => {
 			const handlerWithModel = new ZAiHandler({
 				apiModelId: "glm-4.7",
@@ -507,6 +580,64 @@ describe("ZAiHandler", () => {
 			)
 		})
 
+		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",
+			})
+
+			mockCreate.mockImplementationOnce(() => {
+				return {
+					[Symbol.asyncIterator]: () => ({
+						async next() {
+							return { done: true }
+						},
+					}),
+				}
+			})
+
+			const messageGenerator = handlerWithModel.createMessage("system prompt", [])
+			await messageGenerator.next()
+
+			expect(mockCreate).toHaveBeenCalledWith(
+				expect.objectContaining({
+					model: "glm-5",
+					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",
+			})
+
+			mockCreate.mockImplementationOnce(() => {
+				return {
+					[Symbol.asyncIterator]: () => ({
+						async next() {
+							return { done: true }
+						},
+					}),
+				}
+			})
+
+			const messageGenerator = handlerWithModel.createMessage("system prompt", [])
+			await messageGenerator.next()
+
+			expect(mockCreate).toHaveBeenCalledWith(
+				expect.objectContaining({
+					model: "glm-5",
+					thinking: { type: "disabled" },
+				}),
+			)
+		})
+
 		it("should NOT add thinking parameter for non-thinking models like GLM-4.6", async () => {
 			const handlerWithModel = new ZAiHandler({
 				apiModelId: "glm-4.6",
@@ -532,4 +663,5 @@ describe("ZAiHandler", () => {
 			expect(callArgs.thinking).toBeUndefined()
 		})
 	})
+	// kilocode_change end
 })

src/api/providers/zai.ts

@@ -39,24 +39,25 @@ export class ZAiHandler extends BaseOpenAiCompatibleProvider<string> {
 		})
 	}
 
+	// kilocode_change start
 	/**
-	 * Override createStream to handle GLM-4.7's thinking mode.
-	 * GLM-4.7 has thinking enabled by default in the API, so we need to
-	 * explicitly send { type: "disabled" } when the user turns off reasoning.
+	 * Override createStream to handle Z.ai models with thinking mode.
+	 * Thinking-capable models have reasoning enabled by default in the API,
+	 * so we explicitly send { type: "disabled" } when users turn reasoning off.
 	 */
 	protected override createStream(
 		systemPrompt: string,
 		messages: Anthropic.Messages.MessageParam[],
 		metadata?: ApiHandlerCreateMessageMetadata,
 		requestOptions?: OpenAI.RequestOptions,
 	) {
-		const { id: modelId, info } = this.getModel()
+		const { info } = this.getModel()
 
-		// Check if this is a GLM-4.7 model with thinking support
-		const isThinkingModel = modelId === "glm-4.7" && Array.isArray(info.supportsReasoningEffort)
+		// Thinking models advertise explicit reasoning effort support.
+		const isThinkingModel = Array.isArray(info.supportsReasoningEffort)
 
 		if (isThinkingModel) {
-			// For GLM-4.7, thinking is ON by default in the API.
+			// For thinking-enabled models, thinking is ON by default in the API.
 			// We need to explicitly disable it when reasoning is off.
 			const useReasoning = shouldUseReasoningEffort({ model: info, settings: this.options })
 
@@ -67,9 +68,11 @@ export class ZAiHandler extends BaseOpenAiCompatibleProvider<string> {
 		// For non-thinking models, use the default behavior
 		return super.createStream(systemPrompt, messages, metadata, requestOptions)
 	}
+	// kilocode_change end
 
+	// kilocode_change start
 	/**
-	 * Creates a stream with explicit thinking control for GLM-4.7
+	 * Creates a stream with explicit thinking control for Z.ai thinking models.
 	 */
 	private createStreamWithThinking(
 		systemPrompt: string,
@@ -99,7 +102,7 @@ export class ZAiHandler extends BaseOpenAiCompatibleProvider<string> {
 			messages: [{ role: "system", content: systemPrompt }, ...convertedMessages],
 			stream: true,
 			stream_options: { include_usage: true },
-			// For GLM-4.7: thinking is ON by default, so we explicitly disable when needed
+			// Thinking is ON by default, so we explicitly disable when needed.
 			thinking: useReasoning ? { type: "enabled" } : { type: "disabled" },
 			...(metadata?.tools && { tools: this.convertToolsForOpenAI(metadata.tools) }),
 			...(metadata?.tool_choice && { tool_choice: metadata.tool_choice }),
@@ -110,4 +113,5 @@ export class ZAiHandler extends BaseOpenAiCompatibleProvider<string> {
 
 		return this.client.chat.completions.create(params)
 	}
+	// kilocode_change end
 }

webview-ui/src/components/kilocode/hooks/__tests__/getModelsByProvider.spec.ts

@@ -117,4 +117,11 @@ describe("getOptionsForProvider", () => {
 		const result = getOptionsForProvider("zai", { zaiApiLine: "china_coding" })
 		expect(result).toEqual({ isChina: true })
 	})
+
+	// kilocode_change start
+	it("returns isChina: true for zai provider with china_api apiConfiguration", () => {
+		const result = getOptionsForProvider("zai", { zaiApiLine: "china_api" })
+		expect(result).toEqual({ isChina: true })
+	})
+	// kilocode_change end
 })

webview-ui/src/components/kilocode/hooks/useProviderModels.ts

@@ -357,7 +357,12 @@ export const getOptionsForProvider = (provider: ProviderName, apiConfiguration?:
 	switch (provider) {
 		case "zai":
 			// Determine which Z.AI model set to use based on the API line configuration
-			return { isChina: apiConfiguration?.zaiApiLine === "china_coding" }
+			// kilocode_change start
+			return {
+				isChina:
+					apiConfiguration?.zaiApiLine === "china_coding" || apiConfiguration?.zaiApiLine === "china_api",
+			}
+			// kilocode_change end
 		default:
 			return {}
 	}

webview-ui/src/components/settings/ApiOptions.tsx

@@ -460,7 +460,8 @@ const ApiOptions = ({
 				zai: {
 					field: "apiModelId",
 					default:
-						apiConfiguration.zaiApiLine === "china_coding"
+						// kilocode_change - china_api uses mainland model catalog too.
+						apiConfiguration.zaiApiLine === "china_coding" || apiConfiguration.zaiApiLine === "china_api"
 							? mainlandZAiDefaultModelId
 							: internationalZAiDefaultModelId,
 				},