feat: add GLM-4.7 model with preserved thinking support for Z.ai provider

Author: roomote

packages/types/src/providers/zai.ts

@@ -5,6 +5,8 @@ import { ZaiApiLine } from "../provider-settings.js"
 // https://docs.z.ai/guides/llm/glm-4-32b-0414-128k
 // 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/capabilities/thinking-mode
 // https://docs.z.ai/guides/overview/pricing
 // https://bigmodel.cn/pricing
 
@@ -107,6 +109,21 @@ export const internationalZAiModels = {
 		description:
 			"GLM-4.6 is Zhipu's newest model with an extended context window of up to 200k tokens, providing enhanced capabilities for processing longer documents and conversations.",
 	},
+	"glm-4.7": {
+		maxTokens: 98_304,
+		contextWindow: 200_000,
+		supportsImages: false,
+		supportsPromptCache: true,
+		supportsNativeTools: true,
+		defaultToolProtocol: "native",
+		preserveReasoning: true,
+		inputPrice: 0.6,
+		outputPrice: 2.2,
+		cacheWritesPrice: 0,
+		cacheReadsPrice: 0.11,
+		description:
+			"GLM-4.7 is Zhipu's latest model with preserved thinking capabilities, allowing reasoning content to be retained across turns for improved coherence and cache efficiency.",
+	},
 	"glm-4-32b-0414-128k": {
 		maxTokens: 98_304,
 		contextWindow: 131_072,
@@ -221,6 +238,21 @@ export const mainlandZAiModels = {
 		description:
 			"GLM-4.6 is Zhipu's newest model with an extended context window of up to 200k tokens, providing enhanced capabilities for processing longer documents and conversations.",
 	},
+	"glm-4.7": {
+		maxTokens: 98_304,
+		contextWindow: 204_800,
+		supportsImages: false,
+		supportsPromptCache: true,
+		supportsNativeTools: true,
+		defaultToolProtocol: "native",
+		preserveReasoning: true,
+		inputPrice: 0.29,
+		outputPrice: 1.14,
+		cacheWritesPrice: 0,
+		cacheReadsPrice: 0.057,
+		description:
+			"GLM-4.7 is Zhipu's latest model with preserved thinking capabilities, allowing reasoning content to be retained across turns for improved coherence and cache efficiency.",
+	},
 } as const satisfies Record<string, ModelInfo>
 
 export const ZAI_DEFAULT_TEMPERATURE = 0.6

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

@@ -6,6 +6,7 @@ import { Anthropic } from "@anthropic-ai/sdk"
 import {
 	type InternationalZAiModelId,
 	type MainlandZAiModelId,
+	type ModelInfo,
 	internationalZAiDefaultModelId,
 	mainlandZAiDefaultModelId,
 	internationalZAiModels,
@@ -96,6 +97,23 @@ describe("ZAiHandler", () => {
 			expect(model.info.maxTokens).toBe(16_384)
 			expect(model.info.contextWindow).toBe(131_072)
 		})
+
+		it("should return GLM-4.7 international model with preserved thinking support", () => {
+			const testModelId: InternationalZAiModelId = "glm-4.7"
+			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])
+			// GLM-4.7 should have preserveReasoning enabled for interleaved thinking mode
+			// This allows reasoning_content to be passed back during tool call continuation
+			expect((model.info as ModelInfo).preserveReasoning).toBe(true)
+			expect(model.info.contextWindow).toBe(200_000)
+			expect(model.info.supportsNativeTools).toBe(true)
+		})
 	})
 
 	describe("China Z AI", () => {
@@ -161,6 +179,23 @@ describe("ZAiHandler", () => {
 			expect(model.info.maxTokens).toBe(16_384)
 			expect(model.info.contextWindow).toBe(131_072)
 		})
+
+		it("should return GLM-4.7 China model with preserved thinking support", () => {
+			const testModelId: MainlandZAiModelId = "glm-4.7"
+			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])
+			// GLM-4.7 should have preserveReasoning enabled for interleaved thinking mode
+			// This allows reasoning_content to be passed back during tool call continuation
+			expect((model.info as ModelInfo).preserveReasoning).toBe(true)
+			expect(model.info.contextWindow).toBe(204_800)
+			expect(model.info.supportsNativeTools).toBe(true)
+		})
 	})
 
 	describe("International API", () => {