refactor: use decimal-safe strings for model pricing

apps/code/src/components/CodingModelsShowcase.tsx

@@ -135,7 +135,7 @@ export function CodingModelsShowcase({
 											{provider.inputPrice !== undefined && (
 												<>
 													<span className="font-mono font-medium text-foreground">
-														${formatPrice(provider.inputPrice)}
+														${formatPrice(Number(provider.inputPrice))}
 													</span>
 													<span className="text-muted-foreground"> in</span>
 												</>
@@ -144,7 +144,7 @@ export function CodingModelsShowcase({
 												<>
 													<span className="text-muted-foreground mx-1">/</span>
 													<span className="font-mono font-medium text-foreground">
-														${formatPrice(provider.outputPrice)}
+														${formatPrice(Number(provider.outputPrice))}
 													</span>
 													<span className="text-muted-foreground"> out</span>
 												</>

apps/gateway/src/audio.e2e.ts

@@ -168,7 +168,8 @@ describe("e2e audio input", getTestOptions(), () => {
 			expect(typeof audioInputCost).toBe("number");
 
 			const expected =
-				audioTokens * (provider.inputAudioPrice ?? provider.inputPrice ?? 0);
+				audioTokens *
+				Number(provider.inputAudioPrice ?? provider.inputPrice ?? "0");
 			expect(audioInputCost).toBeCloseTo(expected, 8);
 		},
 	);

apps/gateway/src/chat-prompt-caching.e2e.ts

@@ -270,8 +270,8 @@ describe("e2e prompt caching", getConcurrentTestOptions(), () => {
 
 				if (providerMapping?.cachedInputPrice && providerMapping?.inputPrice) {
 					// Verify the pricing ratio is correct (cached should be cheaper)
-					expect(providerMapping.cachedInputPrice).toBeLessThan(
-						providerMapping.inputPrice,
+					expect(Number(providerMapping.cachedInputPrice)).toBeLessThan(
+						Number(providerMapping.inputPrice),
 					);
 				}
 			},

apps/gateway/src/chat/chat.ts

@@ -1838,7 +1838,9 @@ chat.openapi(completions, async (c) => {
 				// Find the cheapest among the suitable providers for this model
 				for (const provider of suitableProviders) {
 					const totalPrice =
-						((provider.inputPrice ?? 0) + (provider.outputPrice ?? 0)) / 2;
+						(Number(provider.inputPrice ?? "0") +
+							Number(provider.outputPrice ?? "0")) /
+						2;
 
 					if (totalPrice < lowestPrice) {
 						lowestPrice = totalPrice;
@@ -2276,8 +2278,8 @@ chat.openapi(completions, async (c) => {
 							(p) => p.providerId === requestedProvider,
 						);
 						const originalProviderPrice = originalProviderInfo
-							? (originalProviderInfo.inputPrice ?? 0) +
-								(originalProviderInfo.outputPrice ?? 0)
+							? Number(originalProviderInfo.inputPrice ?? "0") +
+								Number(originalProviderInfo.outputPrice ?? "0")
 							: 0;
 
 						const originalProviderScore = {
@@ -2467,8 +2469,8 @@ chat.openapi(completions, async (c) => {
 								(p) => p.providerId === requestedProvider,
 							);
 							const originalProviderPrice = originalProviderInfo
-								? (originalProviderInfo.inputPrice ?? 0) +
-									(originalProviderInfo.outputPrice ?? 0)
+								? Number(originalProviderInfo.inputPrice ?? "0") +
+									Number(originalProviderInfo.outputPrice ?? "0")
 								: 0;
 
 							// Create score entry for the original requested provider
@@ -2691,8 +2693,8 @@ chat.openapi(completions, async (c) => {
 									providerId: rlProviderId,
 									score: -1,
 									price: providerInfo
-										? (providerInfo.inputPrice ?? 0) +
-											(providerInfo.outputPrice ?? 0)
+										? Number(providerInfo.inputPrice ?? "0") +
+											Number(providerInfo.outputPrice ?? "0")
 										: 0,
 									rate_limited: true,
 								});
@@ -2889,8 +2891,8 @@ chat.openapi(completions, async (c) => {
 				{
 					providerId: "custom" as const,
 					modelName: usedModel,
-					inputPrice: 0,
-					outputPrice: 0,
+					inputPrice: "0",
+					outputPrice: "0",
 					contextSize: 8192,
 					maxOutput: 4096,
 					streaming: true,

apps/gateway/src/chat/tools/is-model-truly-free.ts

@@ -9,9 +9,9 @@ export function isModelTrulyFree(modelInfo: ModelDefinition): boolean {
 	}
 	return !modelInfo.providers.some((provider) => {
 		const hasRequestPrice =
-			provider.requestPrice !== undefined && provider.requestPrice > 0;
+			provider.requestPrice !== undefined && Number(provider.requestPrice) > 0;
 		const hasPerSecondPrice = Object.values(provider.perSecondPrice ?? {}).some(
-			(price) => price > 0,
+			(price) => Number(price) > 0,
 		);
 		return hasRequestPrice || hasPerSecondPrice;
 	});

apps/gateway/src/chat/tools/resolve-model-info.ts

@@ -39,8 +39,8 @@ export function resolveModelInfo(
 				{
 					providerId: "custom" as const,
 					modelName: requestedModel,
-					inputPrice: 0,
-					outputPrice: 0,
+					inputPrice: "0",
+					outputPrice: "0",
 					contextSize: 8192,
 					maxOutput: 4096,
 					streaming: true,

apps/gateway/src/lib/anthropic-pricing.spec.ts

@@ -12,9 +12,10 @@ const LEGACY_RATIO_EXCEPTIONS = new Set(["claude-3-haiku-20240307"]);
 function assertRatio(
 	modelName: string,
 	label: string,
-	actual: number,
+	actualStr: string,
 	expected: number,
 ) {
+	const actual = Number(actualStr);
 	expect(
 		actual,
 		`${modelName} ${label}: expected ${expected} (got ${actual}). If Anthropic's published price diverges from the standard multiplier, add the modelName to LEGACY_RATIO_EXCEPTIONS.`,
@@ -78,7 +79,7 @@ describe("Anthropic model pricing", () => {
 			if (provider.inputPrice === undefined) {
 				return;
 			}
-			const base = provider.inputPrice;
+			const base = Number(provider.inputPrice);
 			if (provider.cacheWriteInputPrice !== undefined) {
 				assertRatio(
 					provider.modelName,
@@ -107,7 +108,7 @@ describe("Anthropic model pricing", () => {
 				if (tier.inputPrice === undefined) {
 					continue;
 				}
-				const tierBase = tier.inputPrice;
+				const tierBase = Number(tier.inputPrice);
 				const label = `tier "${tier.name}"`;
 				if (tier.cacheWriteInputPrice !== undefined) {
 					assertRatio(
@@ -197,7 +198,7 @@ describe("AWS Bedrock Anthropic model pricing", () => {
 			if (provider.inputPrice === undefined) {
 				return;
 			}
-			const base = provider.inputPrice;
+			const base = Number(provider.inputPrice);
 			if (provider.cacheWriteInputPrice !== undefined) {
 				assertRatio(
 					provider.modelName,

apps/gateway/src/lib/coding-models.spec.ts

@@ -10,7 +10,7 @@ const baseProvider: ProviderModelMapping = {
 	streaming: true,
 	tools: true,
 	jsonOutputSchema: true,
-	cachedInputPrice: 0.1 / 1e6,
+	cachedInputPrice: "0.1e-6",
 };
 
 const baseModel: ModelDefinition = {
@@ -22,13 +22,13 @@ const baseModel: ModelDefinition = {
 
 describe("providerSupportsCachedInput", () => {
 	it("returns true when cachedInputPrice is set to a positive number", () => {
-		expect(providerSupportsCachedInput({ cachedInputPrice: 0.1 / 1e6 })).toBe(
+		expect(providerSupportsCachedInput({ cachedInputPrice: "0.1e-6" })).toBe(
 			true,
 		);
 	});
 
 	it("returns true when cachedInputPrice is zero", () => {
-		expect(providerSupportsCachedInput({ cachedInputPrice: 0 })).toBe(true);
+		expect(providerSupportsCachedInput({ cachedInputPrice: "0" })).toBe(true);
 	});
 
 	it("returns false when cachedInputPrice is undefined", () => {

apps/gateway/src/lib/costs.spec.ts

@@ -307,7 +307,7 @@ describe("calculateCosts", () => {
 
 	it("should apply discount when model has discount field", async () => {
 		vi.mocked(mockGetEffectiveDiscount).mockResolvedValueOnce({
-			discount: 0.1,
+			discount: "0.1",
 			source: "global_provider",
 			discountId: "disc-global-openai",
 		});
@@ -327,7 +327,7 @@ describe("calculateCosts", () => {
 			null,
 			"openai",
 			"gpt-4",
-			0,
+			"0",
 			"gpt-4",
 		);
 	});
@@ -622,7 +622,7 @@ describe("calculateCosts", () => {
 		const azureProvider = models
 			.find((m) => m.id === "gpt-image-2")
 			?.providers.find((p) => p.providerId === "azure");
-		const discountMultiplier = 1 - (azureProvider?.discount ?? 0);
+		const discountMultiplier = 1 - Number(azureProvider?.discount ?? "0");
 		const expectedTextInputCost =
 			(promptTokens - reportedImageInputTokens) *
 			(5 / 1e6) *

apps/gateway/src/lib/costs.ts

@@ -34,18 +34,18 @@ export function shouldBillCancelledRequests(): boolean {
  */
 function getPricingForTokenCount(
 	pricingTiers: PricingTier[] | undefined,
-	baseInputPrice: number,
-	baseOutputPrice: number,
-	baseCachedInputPrice: number | undefined,
-	baseCacheWriteInputPrice: number | undefined,
-	baseCacheWriteInputPrice1h: number | undefined,
+	baseInputPrice: string,
+	baseOutputPrice: string,
+	baseCachedInputPrice: string | undefined,
+	baseCacheWriteInputPrice: string | undefined,
+	baseCacheWriteInputPrice1h: string | undefined,
 	promptTokens: number,
 ): {
-	inputPrice: number;
-	outputPrice: number;
-	cachedInputPrice: number | undefined;
-	cacheWriteInputPrice: number | undefined;
-	cacheWriteInputPrice1h: number | undefined;
+	inputPrice: string;
+	outputPrice: string;
+	cachedInputPrice: string | undefined;
+	cacheWriteInputPrice: string | undefined;
+	cacheWriteInputPrice1h: string | undefined;
 	tierName: string | undefined;
 } {
 	if (!pricingTiers || pricingTiers.length === 0) {
@@ -292,8 +292,8 @@ export async function calculateCosts(
 	// Get pricing based on token count (supports tiered pricing)
 	const pricing = getPricingForTokenCount(
 		providerInfo.pricingTiers,
-		providerInfo.inputPrice ?? 0,
-		providerInfo.outputPrice ?? 0,
+		providerInfo.inputPrice ?? "0",
+		providerInfo.outputPrice ?? "0",
 		providerInfo.cachedInputPrice,
 		providerInfo.cacheWriteInputPrice,
 		providerInfo.cacheWriteInputPrice1h,
@@ -316,11 +316,11 @@ export async function calculateCosts(
 		pricing.cacheWriteInputPrice1h !== undefined
 			? new Decimal(pricing.cacheWriteInputPrice1h)
 			: cacheWriteInputPrice;
-	const requestPrice = new Decimal(providerInfo.requestPrice ?? 0);
+	const requestPrice = new Decimal(providerInfo.requestPrice ?? "0");
 
 	// Get effective discount (checks org-specific, global, then hardcoded)
 	// Pass both the root model ID and the provider-specific model name for matching
-	const hardcodedDiscount = providerInfo.discount ?? 0;
+	const hardcodedDiscount = providerInfo.discount ?? "0";
 	const effectiveDiscountResult = await getEffectiveDiscount(
 		organizationId,
 		provider,
@@ -600,7 +600,7 @@ export async function calculateCosts(
 		cachedTokens,
 		cacheWriteTokens,
 		estimatedCost: isEstimated,
-		discount: discount !== 0 ? discount : undefined,
+		discount: Number(discount) !== 0 ? Number(discount) : undefined,
 		pricingTier: pricing.tierName,
 	};
 }

apps/gateway/src/lib/iam.spec.ts

@@ -42,22 +42,22 @@ const threeProviderModel: ModelDefinition = {
 			providerId: "google-vertex",
 			modelName: "test-model-vertex",
 			streaming: true,
-			inputPrice: 0.5,
-			outputPrice: 1.0,
+			inputPrice: "0.5",
+			outputPrice: "1.0",
 		},
 		{
 			providerId: "google-ai-studio",
 			modelName: "test-model-studio",
 			streaming: true,
-			inputPrice: 0.5,
-			outputPrice: 1.0,
+			inputPrice: "0.5",
+			outputPrice: "1.0",
 		},
 		{
 			providerId: "openai",
 			modelName: "test-model-openai",
 			streaming: true,
-			inputPrice: 0.3,
-			outputPrice: 0.6,
+			inputPrice: "0.3",
+			outputPrice: "0.6",
 		},
 	],
 };
@@ -75,8 +75,8 @@ const singleActiveProviderModel: ModelDefinition = {
 			providerId: "google-vertex",
 			modelName: "test-model-vertex",
 			streaming: true,
-			inputPrice: 0.5,
-			outputPrice: 1.0,
+			inputPrice: "0.5",
+			outputPrice: "1.0",
 		},
 	],
 };
@@ -102,15 +102,15 @@ const paidModel: ModelDefinition = {
 			providerId: "openai",
 			modelName: "paid-model-openai",
 			streaming: true,
-			inputPrice: 5.0,
-			outputPrice: 15.0,
+			inputPrice: "5.0",
+			outputPrice: "15.0",
 		},
 		{
 			providerId: "anthropic",
 			modelName: "paid-model-anthropic",
 			streaming: true,
-			inputPrice: 3.0,
-			outputPrice: 10.0,
+			inputPrice: "3.0",
+			outputPrice: "10.0",
 		},
 	],
 };

apps/gateway/src/lib/iam.ts

@@ -223,7 +223,7 @@ async function evaluateRule(
 					if (
 						ruleValue.maxInputPrice !== undefined &&
 						provider.inputPrice &&
-						provider.inputPrice > ruleValue.maxInputPrice
+						Number(provider.inputPrice) > ruleValue.maxInputPrice
 					) {
 						return {
 							allowed: false,
@@ -234,7 +234,7 @@ async function evaluateRule(
 					if (
 						ruleValue.maxOutputPrice !== undefined &&
 						provider.outputPrice &&
-						provider.outputPrice > ruleValue.maxOutputPrice
+						Number(provider.outputPrice) > ruleValue.maxOutputPrice
 					) {
 						return {
 							allowed: false,

apps/gateway/src/mcp/mcp.ts

@@ -826,7 +826,7 @@ function createMcpServer(apiKey: string): McpServer {
 					responseText += `- **Family:** ${modelData.family}\n`;
 					if (
 						modelData.requestPrice !== undefined &&
-						modelData.requestPrice > 0
+						Number(modelData.requestPrice) > 0
 					) {
 						responseText += `- **Price:** $${modelData.requestPrice} per request\n`;
 					}

apps/playground/src/lib/model-utils.ts

@@ -1,20 +1,25 @@
 import type { ApiModel, ApiProvider } from "@/lib/fetch-models";
 
-export function formatPrice(price: number | undefined): string {
+export function formatPrice(price: number | string | undefined): string {
 	// Unknown / missing pricing
 	if (price === undefined) {
 		return "Unknown";
 	}
 
+	const n = typeof price === "string" ? Number(price) : price;
+	if (!Number.isFinite(n)) {
+		return "Unknown";
+	}
+
 	// Explicitly free
-	if (price === 0) {
+	if (n === 0) {
 		return "Free";
 	}
 
 	// All model prices in the catalog are stored as "per token" with values like 2 / 1e6.
 	// For the playground we always want to show an explicit per‑million price (to match the /models UI),
 	// otherwise small numbers like 2e‑6 end up rounded to $0.00/1K.
-	const perMillion = price * 1_000_000;
+	const perMillion = n * 1_000_000;
 	// Show full precision (up to 4 decimals) without trailing zeros
 	const formatted = parseFloat(perMillion.toFixed(4)).toString();
 	return `$${formatted}/1M tokens`;