feat(providers): add anthropic-discount provider

.env.example

@@ -138,6 +138,10 @@ LLM_OPENAI_API_KEY=sk-your_openai_key_here
 # Anthropic
 LLM_ANTHROPIC_API_KEY=sk-ant-your_anthropic_key_here
 
+# Anthropic Discount (Anthropic-compatible endpoint via custom base URL + API key)
+LLM_ANTHROPIC_DISCOUNT_BASE_URL=https://your-anthropic-discount-endpoint
+LLM_ANTHROPIC_DISCOUNT_API_KEY=your_anthropic_discount_key_here
+
 # Google AI Studio
 LLM_GOOGLE_AI_STUDIO_API_KEY=your_google_ai_studio_key_here
 

apps/gateway/src/chat/chat.ts

@@ -4585,7 +4585,11 @@ chat.openapi(completions, async (c) => {
 	}
 
 	// Anthropic does not allow temperature and top_p to be set simultaneously
-	if (usedProvider === "anthropic" || usedProvider === "vertex-anthropic") {
+	if (
+		usedProvider === "anthropic" ||
+		usedProvider === "anthropic-discount" ||
+		usedProvider === "vertex-anthropic"
+	) {
 		if (temperature !== undefined && top_p !== undefined) {
 			top_p = undefined;
 		}
@@ -5270,7 +5274,11 @@ chat.openapi(completions, async (c) => {
 						headers["Content-Type"] = "application/json";
 
 						// Add effort beta header for Anthropic if effort parameter is specified
-						if (usedProvider === "anthropic" && effort !== undefined) {
+						if (
+							(usedProvider === "anthropic" ||
+								usedProvider === "anthropic-discount") &&
+							effort !== undefined
+						) {
 							const currentBeta = headers["anthropic-beta"];
 							headers["anthropic-beta"] = currentBeta
 								? `${currentBeta},effort-2025-11-24`
@@ -5279,7 +5287,8 @@ chat.openapi(completions, async (c) => {
 
 						// Add structured outputs beta header for Anthropic if json_schema response_format is specified
 						if (
-							usedProvider === "anthropic" &&
+							(usedProvider === "anthropic" ||
+								usedProvider === "anthropic-discount") &&
 							response_format?.type === "json_schema"
 						) {
 							const currentBeta = headers["anthropic-beta"];
@@ -6556,6 +6565,7 @@ chat.openapi(completions, async (c) => {
 					streamingIsJsonResponseFormat &&
 					(streamingResponseHealingEnabled === true ||
 						((usedProvider === "anthropic" ||
+							usedProvider === "anthropic-discount" ||
 							usedProvider === "vertex-anthropic") &&
 							response_format?.type === "json_object") ||
 						(usedProvider === "aws-bedrock" &&
@@ -7374,6 +7384,7 @@ chat.openapi(completions, async (c) => {
 								// For Anthropic, if we have partial usage data, complete it
 								if (
 									(usedProvider === "anthropic" ||
+										usedProvider === "anthropic-discount" ||
 										usedProvider === "vertex-anthropic") &&
 									transformedData.usage
 								) {
@@ -7448,6 +7459,7 @@ chat.openapi(completions, async (c) => {
 								// from the initial content_block_start event. This ensures OpenAI SDK compatibility.
 								if (
 									usedProvider === "anthropic" ||
+									usedProvider === "anthropic-discount" ||
 									usedProvider === "vertex-anthropic"
 								) {
 									const toolCalls =
@@ -7577,6 +7589,7 @@ chat.openapi(completions, async (c) => {
 								const contentChunk = extractContent(
 									isGoogleCompatibleProvider(usedProvider) ||
 										usedProvider === "anthropic" ||
+										usedProvider === "anthropic-discount" ||
 										usedProvider === "vertex-anthropic"
 										? data
 										: transformedData,
@@ -7610,6 +7623,7 @@ chat.openapi(completions, async (c) => {
 								// Check for web search results based on provider-specific data
 								if (
 									usedProvider === "anthropic" ||
+									usedProvider === "anthropic-discount" ||
 									usedProvider === "vertex-anthropic"
 								) {
 									// For Anthropic, count web_search_tool_result blocks
@@ -7653,6 +7667,7 @@ chat.openapi(completions, async (c) => {
 								const reasoningContentChunk = extractReasoning(
 									isGoogleCompatibleProvider(usedProvider) ||
 										usedProvider === "anthropic" ||
+										usedProvider === "anthropic-discount" ||
 										usedProvider === "vertex-anthropic"
 										? data
 										: transformedData,
@@ -7682,6 +7697,7 @@ chat.openapi(completions, async (c) => {
 										// For Anthropic content_block_delta events, match by content block index
 										if (
 											(usedProvider === "anthropic" ||
+												usedProvider === "anthropic-discount" ||
 												usedProvider === "vertex-anthropic") &&
 											newCall._contentBlockIndex !== undefined
 										) {
@@ -7727,6 +7743,7 @@ chat.openapi(completions, async (c) => {
 										}
 										break;
 									case "anthropic":
+									case "anthropic-discount":
 									case "vertex-anthropic":
 										if (
 											data.type === "message_delta" &&
@@ -8975,7 +8992,11 @@ chat.openapi(completions, async (c) => {
 			}
 
 			// Add effort beta header for Anthropic if effort parameter is specified
-			if (usedProvider === "anthropic" && effort !== undefined) {
+			if (
+				(usedProvider === "anthropic" ||
+					usedProvider === "anthropic-discount") &&
+				effort !== undefined
+			) {
 				const currentBeta = headers["anthropic-beta"];
 				headers["anthropic-beta"] = currentBeta
 					? `${currentBeta},effort-2025-11-24`
@@ -8984,7 +9005,8 @@ chat.openapi(completions, async (c) => {
 
 			// Add structured outputs beta header for Anthropic if json_schema response_format is specified
 			if (
-				usedProvider === "anthropic" &&
+				(usedProvider === "anthropic" ||
+					usedProvider === "anthropic-discount") &&
 				response_format?.type === "json_schema"
 			) {
 				const currentBeta = headers["anthropic-beta"];
@@ -10337,7 +10359,9 @@ chat.openapi(completions, async (c) => {
 	const shouldHealNonStreaming =
 		isJsonResponseFormat &&
 		(responseHealingEnabled === true ||
-			((usedProvider === "anthropic" || usedProvider === "vertex-anthropic") &&
+			((usedProvider === "anthropic" ||
+				usedProvider === "anthropic-discount" ||
+				usedProvider === "vertex-anthropic") &&
 				response_format?.type === "json_object") ||
 			(usedProvider === "aws-bedrock" &&
 				response_format?.type === "json_object") ||

apps/gateway/src/chat/tools/extract-content.ts

@@ -14,6 +14,7 @@ export function extractContent(data: any, provider: Provider): string {
 			return contentParts.map((part: any) => part.text).join("") ?? "";
 		}
 		case "anthropic":
+		case "anthropic-discount":
 		case "vertex-anthropic":
 			if (data.type === "content_block_delta" && data.delta?.text) {
 				return data.delta.text;

apps/gateway/src/chat/tools/extract-reasoning.ts

@@ -34,6 +34,7 @@ export function extractReasoning(
 ): string {
 	switch (provider) {
 		case "anthropic":
+		case "anthropic-discount":
 		case "vertex-anthropic": {
 			const chunk = data as AnthropicStreamChunk;
 			if (

apps/gateway/src/chat/tools/extract-token-usage.ts

@@ -162,6 +162,7 @@ export function extractTokenUsage(
 			}
 			break;
 		case "anthropic":
+		case "anthropic-discount":
 			{
 				const usage = data.message?.usage ?? data.usage;
 				if (!usage) {

apps/gateway/src/chat/tools/extract-tool-calls.ts

@@ -11,6 +11,7 @@ export function extractToolCalls(
 ): any[] | null {
 	switch (provider) {
 		case "anthropic":
+		case "anthropic-discount":
 		case "vertex-anthropic":
 			// Anthropic streaming tool calls come as content_block_start with tool_use type
 			if (

apps/gateway/src/chat/tools/map-finish-reason-to-openai.ts

@@ -62,6 +62,7 @@ export function mapFinishReasonToOpenai(
 					return "stop";
 			}
 		case "anthropic":
+		case "anthropic-discount":
 		case "vertex-anthropic":
 			if (!finishReason) {
 				return hasToolCalls ? "tool_calls" : "stop";

apps/gateway/src/chat/tools/parse-provider-response.ts

@@ -138,6 +138,7 @@ export function parseProviderResponse(
 			break;
 		}
 		case "anthropic":
+		case "anthropic-discount":
 		case "vertex-anthropic": {
 			// Extract content and reasoning content from Anthropic response
 			const contentBlocks = json.content ?? [];

apps/gateway/src/chat/tools/resolve-provider-context.ts

@@ -421,7 +421,11 @@ export async function resolveProviderContext(
 	}
 
 	// Anthropic does not allow temperature and top_p simultaneously
-	if (usedProvider === "anthropic" || usedProvider === "vertex-anthropic") {
+	if (
+		usedProvider === "anthropic" ||
+		usedProvider === "anthropic-discount" ||
+		usedProvider === "vertex-anthropic"
+	) {
 		if (temperature !== undefined && top_p !== undefined) {
 			top_p = undefined;
 		}
@@ -516,15 +520,18 @@ export async function resolveProviderContext(
 	});
 	headers["Content-Type"] = "application/json";
 
-	if (usedProvider === "anthropic" && options.effort !== undefined) {
+	if (
+		(usedProvider === "anthropic" || usedProvider === "anthropic-discount") &&
+		options.effort !== undefined
+	) {
 		const currentBeta = headers["anthropic-beta"];
 		headers["anthropic-beta"] = currentBeta
 			? `${currentBeta},effort-2025-11-24`
 			: "effort-2025-11-24";
 	}
 
 	if (
-		usedProvider === "anthropic" &&
+		(usedProvider === "anthropic" || usedProvider === "anthropic-discount") &&
 		options.response_format?.type === "json_schema"
 	) {
 		const currentBeta = headers["anthropic-beta"];

apps/gateway/src/chat/tools/transform-response-to-openai.ts

@@ -425,6 +425,7 @@ export function transformResponseToOpenai(
 			break;
 		}
 		case "anthropic":
+		case "anthropic-discount":
 		case "vertex-anthropic": {
 			transformedResponse = {
 				id: `chatcmpl-${Date.now()}`,

apps/gateway/src/chat/tools/transform-streaming-to-openai.ts

@@ -89,6 +89,7 @@ export function transformStreamingToOpenai(
 
 	switch (usedProvider) {
 		case "anthropic":
+		case "anthropic-discount":
 		case "vertex-anthropic": {
 			const usage = data.message?.usage ?? data.usage;
 			if (data.type === "message_start") {

apps/gateway/src/lib/logs.ts

@@ -66,6 +66,7 @@ export function getUnifiedFinishReason(
 
 	switch (provider) {
 		case "anthropic":
+		case "anthropic-discount":
 		case "vertex-anthropic":
 			if (finishReason === "stop_sequence") {
 				return UnifiedFinishReason.COMPLETED;

packages/actions/src/get-provider-endpoint.ts

@@ -132,6 +132,16 @@ export function getProviderEndpoint(
 			case "anthropic":
 				url = "https://api.anthropic.com";
 				break;
+			case "anthropic-discount":
+				url = skipEnvVars
+					? undefined
+					: getProviderEnvValue("anthropic-discount", "baseUrl", configIndex);
+				if (!url) {
+					throw new Error(
+						"Anthropic Discount provider requires LLM_ANTHROPIC_DISCOUNT_BASE_URL environment variable",
+					);
+				}
+				break;
 			case "google-ai-studio":
 				url =
 					envValueOrDefault(
@@ -337,6 +347,7 @@ export function getProviderEndpoint(
 
 	switch (provider) {
 		case "anthropic":
+		case "anthropic-discount":
 			return `${url}/v1/messages`;
 		case "google-ai-studio": {
 			const endpoint = stream ? "streamGenerateContent" : "generateContent";

packages/actions/src/get-provider-headers.ts

@@ -22,7 +22,8 @@ export function getProviderHeaders(
 	}
 
 	switch (provider) {
-		case "anthropic": {
+		case "anthropic":
+		case "anthropic-discount": {
 			const betaFeatures = ["tools-2024-04-04", "prompt-caching-2024-07-31"];
 			if (options?.webSearchEnabled) {
 				betaFeatures.push("web-search-2025-03-05");

packages/actions/src/prepare-request-body.ts

@@ -1040,6 +1040,7 @@ export async function prepareRequestBody(
 	// Anthropic-style `ttl: "1h"` must be normalized away before forwarding.
 	const providerHandlesCacheControl =
 		usedProvider === "anthropic" ||
+		usedProvider === "anthropic-discount" ||
 		usedProvider === "vertex-anthropic" ||
 		usedProvider === "aws-bedrock" ||
 		usedProvider === "alibaba";
@@ -1506,6 +1507,7 @@ export async function prepareRequestBody(
 			break;
 		}
 		case "anthropic":
+		case "anthropic-discount":
 		case "vertex-anthropic": {
 			// Remove generic tool_choice that was added earlier
 			delete requestBody.tool_choice;

packages/actions/src/transform-anthropic-messages.ts

@@ -35,7 +35,8 @@ export async function transformAnthropicMessages(
 	// Apply for anthropic provider only, and only when the project hasn't
 	// opted out of auto-injection.
 	const shouldApplyCacheControl =
-		provider === "anthropic" && providerCacheControlEnabled;
+		(provider === "anthropic" || provider === "anthropic-discount") &&
+		providerCacheControlEnabled;
 
 	// Track cache_control usage to limit to maximum of 4 blocks total (including system messages)
 	let cacheControlCount = initialCacheControlCount;

packages/models/src/models/anthropic.ts

@@ -1058,6 +1058,30 @@ export const anthropicModels = [
 				webSearch: true,
 				webSearchPrice: "0.01", // $10 per 1000 searches
 			},
+			{
+				providerId: "anthropic-discount",
+				externalId: "claude-opus-4-7",
+				inputPrice: "5.0e-6",
+				outputPrice: "25.0e-6",
+				cachedInputPrice: "0.5e-6",
+				cacheWriteInputPrice: "6.25e-6",
+				cacheWriteInputPrice1h: "10.0e-6",
+				discount: "0.1",
+				minCacheableTokens: 4096,
+				requestPrice: "0",
+				contextSize: 1000000,
+				maxOutput: 128000,
+				reasoning: true,
+				reasoningMode: "adaptive",
+				reasoningOutput: "omit",
+				streaming: true,
+				vision: true,
+				tools: true,
+				jsonOutputSchema: true,
+				supportedParameters: ["temperature", "max_tokens", "top_p", "effort"],
+				webSearch: true,
+				webSearchPrice: "0.01",
+			},
 			{
 				providerId: "aws-bedrock",
 				externalId: "anthropic.claude-opus-4-7",
@@ -1130,6 +1154,30 @@ export const anthropicModels = [
 				webSearch: true,
 				webSearchPrice: "0.01",
 			},
+			{
+				providerId: "anthropic-discount",
+				externalId: "claude-opus-4-8",
+				inputPrice: "5.0e-6",
+				outputPrice: "25.0e-6",
+				cachedInputPrice: "0.5e-6",
+				cacheWriteInputPrice: "6.25e-6",
+				cacheWriteInputPrice1h: "10.0e-6",
+				discount: "0.1",
+				minCacheableTokens: 4096,
+				requestPrice: "0",
+				contextSize: 1000000,
+				maxOutput: 128000,
+				reasoning: true,
+				reasoningMode: "adaptive",
+				reasoningOutput: "omit",
+				streaming: true,
+				vision: true,
+				tools: true,
+				jsonOutputSchema: true,
+				supportedParameters: ["temperature", "max_tokens", "top_p", "effort"],
+				webSearch: true,
+				webSearchPrice: "0.01",
+			},
 		],
 	},
 ] as const satisfies ModelDefinition[];