feat: add thinking mode support for GLM-4.6 and GLM-4.6V models

- Add supportsReasoningEffort to GLM-4.6 and GLM-4.6V model definitions
- Update GLM detection logic to recognize GLM-4.6/4.6V thinking support
- Update Z.ai provider to handle thinking mode for all GLM models with reasoning
- Update tests to reflect GLM-4.6/4.6V thinking support

Addresses issue #11071 where Z.ai documentation shows GLM-4.6 and GLM-4.6V
support thinking mode, but this was not reflected in the codebase.

packages/types/src/providers/zai.ts

@@ -86,6 +86,9 @@ export const internationalZAiModels = {
 		contextWindow: 131_072,
 		supportsImages: true,
 		supportsPromptCache: true,
+		supportsReasoningEffort: ["disable", "medium"],
+		reasoningEffort: "medium",
+		preserveReasoning: true,
 		inputPrice: 0.3,
 		outputPrice: 0.9,
 		cacheWritesPrice: 0,
@@ -98,6 +101,9 @@ export const internationalZAiModels = {
 		contextWindow: 200_000,
 		supportsImages: false,
 		supportsPromptCache: true,
+		supportsReasoningEffort: ["disable", "medium"],
+		reasoningEffort: "medium",
+		preserveReasoning: true,
 		inputPrice: 0.6,
 		outputPrice: 2.2,
 		cacheWritesPrice: 0,
@@ -259,6 +265,9 @@ export const mainlandZAiModels = {
 		contextWindow: 204_800,
 		supportsImages: false,
 		supportsPromptCache: true,
+		supportsReasoningEffort: ["disable", "medium"],
+		reasoningEffort: "medium",
+		preserveReasoning: true,
 		inputPrice: 0.29,
 		outputPrice: 1.14,
 		cacheWritesPrice: 0,
@@ -310,6 +319,9 @@ export const mainlandZAiModels = {
 		contextWindow: 131_072,
 		supportsImages: true,
 		supportsPromptCache: true,
+		supportsReasoningEffort: ["disable", "medium"],
+		reasoningEffort: "medium",
+		preserveReasoning: true,
 		inputPrice: 0.15,
 		outputPrice: 0.45,
 		cacheWritesPrice: 0,

src/api/providers/utils/__tests__/glm-model-detection.spec.ts

@@ -130,17 +130,28 @@ describe("GLM Model Detection", () => {
 		})
 
 		describe("thinking support detection", () => {
-			it("should detect thinking support for GLM-4.7", () => {
+			it("should detect thinking support for GLM-4.7 variants", () => {
 				expect(detectGlmModel("glm-4.7").supportsThinking).toBe(true)
 				expect(detectGlmModel("glm-4.7-flash").supportsThinking).toBe(true)
 				expect(detectGlmModel("GLM-4.7-FlashX").supportsThinking).toBe(true)
 			})
 
-			it("should NOT detect thinking support for GLM-4.5 and GLM-4.6", () => {
+			it("should detect thinking support for GLM-4.6 base model", () => {
+				expect(detectGlmModel("glm-4.6").supportsThinking).toBe(true)
+			})
+
+			it("should detect thinking support for GLM-4.6V vision variants", () => {
+				expect(detectGlmModel("glm-4.6v").supportsThinking).toBe(true)
+				expect(detectGlmModel("GLM-4.6V").supportsThinking).toBe(true)
+				expect(detectGlmModel("glm-4.6v-flash").supportsThinking).toBe(true)
+				expect(detectGlmModel("glm-4.6v-flashx").supportsThinking).toBe(true)
+			})
+
+			it("should NOT detect thinking support for GLM-4.5 variants", () => {
 				expect(detectGlmModel("glm-4.5").supportsThinking).toBe(false)
-				expect(detectGlmModel("glm-4.6").supportsThinking).toBe(false)
 				expect(detectGlmModel("glm-4.5-air").supportsThinking).toBe(false)
-				expect(detectGlmModel("glm-4.6v").supportsThinking).toBe(false)
+				expect(detectGlmModel("glm-4.5-flash").supportsThinking).toBe(false)
+				expect(detectGlmModel("glm-4.5v").supportsThinking).toBe(false)
 			})
 		})
 

src/api/providers/utils/glm-model-detection.ts

@@ -138,8 +138,12 @@ export function detectGlmModel(modelId: string): GlmModelConfig {
 		variant = "x"
 	}
 
-	// GLM-4.7 has built-in thinking support
-	const supportsThinking = version === "4.7"
+	// GLM-4.6, GLM-4.6V, and GLM-4.7 have built-in thinking support
+	// For GLM-4.6, only the base model and vision variants support thinking
+	const supportsThinking =
+		version === "4.7" ||
+		(version === "4.6" &&
+			(variant === "base" || variant === "v" || variant === "v-flash" || variant === "v-flashx"))
 
 	// Generate display name
 	let displayName = `GLM-${version !== "unknown" ? version : "4.x"}`

src/api/providers/zai.ts

@@ -40,9 +40,9 @@ export class ZAiHandler extends BaseOpenAiCompatibleProvider<string> {
 	}
 
 	/**
-	 * 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 thinking mode for GLM models.
+	 * GLM-4.6, GLM-4.6V, and GLM-4.7 have thinking enabled by default in the API,
+	 * so we need to explicitly send { type: "disabled" } when the user turns off reasoning.
 	 */
 	protected override createStream(
 		systemPrompt: string,
@@ -52,11 +52,11 @@ export class ZAiHandler extends BaseOpenAiCompatibleProvider<string> {
 	) {
 		const { id: modelId, 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)
+		// Check if this is a GLM model with thinking support (GLM-4.6, GLM-4.6V, GLM-4.7)
+		const isThinkingModel = Array.isArray(info.supportsReasoningEffort)
 
 		if (isThinkingModel) {
-			// For GLM-4.7, thinking is ON by default in the API.
+			// For GLM thinking 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 })
 
@@ -69,7 +69,7 @@ export class ZAiHandler extends BaseOpenAiCompatibleProvider<string> {
 	}
 
 	/**
-	 * Creates a stream with explicit thinking control for GLM-4.7
+	 * Creates a stream with explicit thinking control for GLM thinking models (4.6, 4.6V, 4.7)
 	 */
 	private createStreamWithThinking(
 		systemPrompt: string,
@@ -99,7 +99,7 @@ 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
+			// For GLM thinking models: thinking is ON by default, so we explicitly disable when needed
 			thinking: useReasoning ? { type: "enabled" } : { type: "disabled" },
 			tools: this.convertToolsForOpenAI(metadata?.tools),
 			tool_choice: metadata?.tool_choice,