feat: add OpenRouter quantization filter to exclude low-bit providers (rebased)

packages/types/src/provider-settings.ts

@@ -206,6 +206,7 @@ const openRouterSchema = baseProviderSettingsSchema.extend({
 	openRouterModelId: z.string().optional(),
 	openRouterBaseUrl: z.string().optional(),
 	openRouterSpecificProvider: z.string().optional(),
+	openRouterExcludeLowQuantization: z.boolean().optional(),
 })
 
 const bedrockSchema = apiModelIdProviderModelSchema.extend({

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

@@ -698,4 +698,77 @@ describe("OpenRouterHandler", () => {
 			)
 		})
 	})
+
+	describe("buildProviderOptions", () => {
+		it("returns undefined when no specific provider and no quantization filter", () => {
+			const handler = new OpenRouterHandler({
+				openRouterApiKey: "test-key",
+				openRouterModelId: "anthropic/claude-sonnet-4",
+			})
+			const result = (handler as any).buildProviderOptions()
+			expect(result).toBeUndefined()
+		})
+
+		it("returns provider routing when specific provider is set", () => {
+			const handler = new OpenRouterHandler({
+				openRouterApiKey: "test-key",
+				openRouterModelId: "anthropic/claude-sonnet-4",
+				openRouterSpecificProvider: "Anthropic",
+			})
+			const result = (handler as any).buildProviderOptions()
+			expect(result).toEqual({
+				order: ["Anthropic"],
+				only: ["Anthropic"],
+				allow_fallbacks: false,
+			})
+		})
+
+		it("returns quantizations when excludeLowQuantization is enabled", () => {
+			const handler = new OpenRouterHandler({
+				openRouterApiKey: "test-key",
+				openRouterModelId: "anthropic/claude-sonnet-4",
+				openRouterExcludeLowQuantization: true,
+			})
+			const result = (handler as any).buildProviderOptions()
+			expect(result).toEqual({
+				quantizations: ["fp16", "bf16", "fp8", "int8"],
+			})
+		})
+
+		it("combines specific provider and quantization filter", () => {
+			const handler = new OpenRouterHandler({
+				openRouterApiKey: "test-key",
+				openRouterModelId: "anthropic/claude-sonnet-4",
+				openRouterSpecificProvider: "Anthropic",
+				openRouterExcludeLowQuantization: true,
+			})
+			const result = (handler as any).buildProviderOptions()
+			expect(result).toEqual({
+				order: ["Anthropic"],
+				only: ["Anthropic"],
+				allow_fallbacks: false,
+				quantizations: ["fp16", "bf16", "fp8", "int8"],
+			})
+		})
+
+		it("returns undefined when specific provider is the default", () => {
+			const handler = new OpenRouterHandler({
+				openRouterApiKey: "test-key",
+				openRouterModelId: "anthropic/claude-sonnet-4",
+				openRouterSpecificProvider: "[default]",
+			})
+			const result = (handler as any).buildProviderOptions()
+			expect(result).toBeUndefined()
+		})
+
+		it("returns undefined when excludeLowQuantization is false", () => {
+			const handler = new OpenRouterHandler({
+				openRouterApiKey: "test-key",
+				openRouterModelId: "anthropic/claude-sonnet-4",
+				openRouterExcludeLowQuantization: false,
+			})
+			const result = (handler as any).buildProviderOptions()
+			expect(result).toBeUndefined()
+		})
+	})
 })

src/api/providers/openrouter.ts

@@ -317,15 +317,7 @@ export class OpenRouterHandler extends BaseProvider implements SingleCompletionH
 			messages: openAiMessages,
 			stream: true,
 			stream_options: { include_usage: true },
-			// Only include provider if openRouterSpecificProvider is not "[default]".
-			...(this.options.openRouterSpecificProvider &&
-				this.options.openRouterSpecificProvider !== OPENROUTER_DEFAULT_PROVIDER_NAME && {
-					provider: {
-						order: [this.options.openRouterSpecificProvider],
-						only: [this.options.openRouterSpecificProvider],
-						allow_fallbacks: false,
-					},
-				}),
+			...(this.buildProviderOptions() && { provider: this.buildProviderOptions() }),
 			...(reasoning && { reasoning }),
 			tools: this.convertToolsForOpenAI(metadata?.tools),
 			tool_choice: metadata?.tool_choice,
@@ -574,6 +566,47 @@ export class OpenRouterHandler extends BaseProvider implements SingleCompletionH
 		return { id, info, topP: isDeepSeekR1 ? 0.95 : undefined, ...params }
 	}
 
+	/**
+	 * Build the `provider` options object for OpenRouter requests.
+	 * Combines specific provider routing and quantization filtering.
+	 */
+	private buildProviderOptions():
+		| {
+				order?: string[]
+				only?: string[]
+				allow_fallbacks?: boolean
+				quantizations?: string[]
+		  }
+		| undefined {
+		const hasSpecificProvider =
+			this.options.openRouterSpecificProvider &&
+			this.options.openRouterSpecificProvider !== OPENROUTER_DEFAULT_PROVIDER_NAME
+		const excludeLowQuantization = this.options.openRouterExcludeLowQuantization
+
+		if (!hasSpecificProvider && !excludeLowQuantization) {
+			return undefined
+		}
+
+		const provider: {
+			order?: string[]
+			only?: string[]
+			allow_fallbacks?: boolean
+			quantizations?: string[]
+		} = {}
+
+		if (hasSpecificProvider) {
+			provider.order = [this.options.openRouterSpecificProvider!]
+			provider.only = [this.options.openRouterSpecificProvider!]
+			provider.allow_fallbacks = false
+		}
+
+		if (excludeLowQuantization) {
+			provider.quantizations = ["fp16", "bf16", "fp8", "int8"]
+		}
+
+		return provider
+	}
+
 	async completePrompt(prompt: string) {
 		let { id: modelId, maxTokens, temperature, reasoning } = await this.fetchModel()
 
@@ -583,15 +616,7 @@ export class OpenRouterHandler extends BaseProvider implements SingleCompletionH
 			temperature,
 			messages: [{ role: "user", content: prompt }],
 			stream: false,
-			// Only include provider if openRouterSpecificProvider is not "[default]".
-			...(this.options.openRouterSpecificProvider &&
-				this.options.openRouterSpecificProvider !== OPENROUTER_DEFAULT_PROVIDER_NAME && {
-					provider: {
-						order: [this.options.openRouterSpecificProvider],
-						only: [this.options.openRouterSpecificProvider],
-						allow_fallbacks: false,
-					},
-				}),
+			...(this.buildProviderOptions() && { provider: this.buildProviderOptions() }),
 			...(reasoning && { reasoning }),
 		}
 

webview-ui/src/components/settings/providers/OpenRouter.tsx

@@ -103,6 +103,18 @@ export const OpenRouter = ({
 					)}
 				</div>
 			)}
+			<div>
+				<Checkbox
+					checked={apiConfiguration?.openRouterExcludeLowQuantization ?? false}
+					onChange={(checked: boolean) => {
+						setApiConfigurationField("openRouterExcludeLowQuantization", checked)
+					}}>
+					{t("settings:providers.openRouter.excludeLowQuantization.label")}
+				</Checkbox>
+				<div className="text-sm text-vscode-descriptionForeground mt-1 ml-6">
+					{t("settings:providers.openRouter.excludeLowQuantization.description")}
+				</div>
+			</div>
 			<ModelPicker
 				apiConfiguration={apiConfiguration}
 				setApiConfigurationField={setApiConfigurationField}

webview-ui/src/i18n/locales/en/settings.json

@@ -540,6 +540,10 @@
 				"title": "OpenRouter Provider Routing",
 				"description": "OpenRouter routes requests to the best available providers for your model. By default, requests are load balanced across the top providers to maximize uptime. However, you can choose a specific provider to use for this model.",
 				"learnMore": "Learn more about provider routing"
+			},
+			"excludeLowQuantization": {
+				"label": "Exclude low-bit quantization (FP4/FP6/Int4)",
+				"description": "Only allow higher precision providers (FP8/FP16/BF16/Int8). Helps prevent broken CJK (Korean/Chinese/Japanese) encoding from aggressively quantized models."
 			}
 		},
 		"customModel": {