reproduction (reproduced)

Author: lgrammel

examples/ai-functions/src/reproduction/issue-15894-ai-gateway-deepseek-v32-metadata.ts

@@ -0,0 +1,153 @@
+type GatewayPricing = {
+  input?: string;
+  output?: string;
+  prompt?: string;
+  completion?: string;
+  input_cache_read?: string;
+};
+
+type GatewayModel = {
+  id: string;
+  owned_by?: string;
+  max_tokens?: number;
+  pricing?: GatewayPricing;
+};
+
+type GatewayEndpoint = {
+  provider_name: string;
+  max_completion_tokens?: number;
+  pricing?: GatewayPricing;
+};
+
+type GatewayEndpointsResponse = {
+  data: {
+    id: string;
+    endpoints: GatewayEndpoint[];
+  };
+};
+
+const gatewayModelsUrl = 'https://ai-gateway.vercel.sh/v1/models';
+const gatewayModelPageBaseUrl = 'https://vercel.com/ai-gateway/models';
+
+async function fetchJson<T>(url: string): Promise<T> {
+  const response = await fetch(url);
+
+  if (!response.ok) {
+    throw new Error(`GET ${url} failed with ${response.status}`);
+  }
+
+  return response.json() as Promise<T>;
+}
+
+async function fetchText(url: string): Promise<string> {
+  const response = await fetch(url);
+
+  if (!response.ok) {
+    throw new Error(`GET ${url} failed with ${response.status}`);
+  }
+
+  return response.text();
+}
+
+function stripHtml(html: string): string {
+  return html
+    .replace(/<script[\s\S]*?<\/script>/gi, ' ')
+    .replace(/<style[\s\S]*?<\/style>/gi, ' ')
+    .replace(/<[^>]+>/g, ' ')
+    .replace(/&nbsp;/g, ' ')
+    .replace(/&amp;/g, '&')
+    .replace(/&#x27;/g, "'")
+    .replace(/&quot;/g, '"')
+    .replace(/\s+/g, ' ')
+    .trim();
+}
+
+function hasPrice(
+  pricing: GatewayPricing | undefined,
+  input: string,
+  output: string,
+): boolean {
+  return (
+    (pricing?.input === input || pricing?.prompt === input) &&
+    (pricing?.output === output || pricing?.completion === output)
+  );
+}
+
+async function main() {
+  const failures: string[] = [];
+  const modelsResponse = await fetchJson<{ data: GatewayModel[] }>(
+    gatewayModelsUrl,
+  );
+  const thinkingModelId = 'deepseek/deepseek-v3.2-thinking';
+  const thinkingModel = modelsResponse.data.find(
+    model => model.id === thinkingModelId,
+  );
+
+  if (thinkingModel == null) {
+    throw new Error(
+      `${thinkingModelId} was not present in ${gatewayModelsUrl}`,
+    );
+  }
+
+  const endpointsUrl = `${gatewayModelsUrl}/${thinkingModelId}/endpoints`;
+  const endpointsResponse =
+    await fetchJson<GatewayEndpointsResponse>(endpointsUrl);
+  const endpointProviders = endpointsResponse.data.endpoints.map(
+    endpoint => endpoint.provider_name,
+  );
+  const bedrockEndpoint = endpointsResponse.data.endpoints.find(
+    endpoint => endpoint.provider_name === 'bedrock',
+  );
+
+  const pageUrl = `${gatewayModelPageBaseUrl}/deepseek-v3.2-thinking`;
+  const pageText = stripHtml(await fetchText(pageUrl));
+
+  const pageClaimsDeepSeekProvider = pageText.includes(
+    'The Thinking variant and standard V3.2 are accessible through AI Gateway under the deepseek provider',
+  );
+  if (pageClaimsDeepSeekProvider && !endpointProviders.includes('deepseek')) {
+    failures.push(
+      `${pageUrl} says the Thinking variant is available under the deepseek provider, but ${endpointsUrl} only lists providers: ${endpointProviders.join(
+        ', ',
+      )}.`,
+    );
+  }
+
+  if (
+    bedrockEndpoint != null &&
+    hasPrice(thinkingModel.pricing, '0.00000062', '0.00000185') &&
+    hasPrice(bedrockEndpoint.pricing, '0.00000062', '0.00000185') &&
+    !('pricing_provider' in thinkingModel)
+  ) {
+    failures.push(
+      `${gatewayModelsUrl} exposes aggregate pricing for ${thinkingModelId} as Bedrock-like $0.62/M input and $1.85/M output, but the model object does not identify Bedrock as the aggregate pricing source.`,
+    );
+  }
+
+  if (
+    thinkingModel.max_tokens === 8000 &&
+    pageText.includes('generates up to 163K tokens') &&
+    pageText.includes('The output token budget extends to 163K tokens')
+  ) {
+    failures.push(
+      `${gatewayModelsUrl} reports max_tokens=8000 for ${thinkingModelId}, while ${pageUrl} says the model generates up to 163K tokens and has a 163K output token budget.`,
+    );
+  }
+
+  if (failures.length > 0) {
+    throw new Error(
+      `AI Gateway DeepSeek V3.2 catalog metadata is inconsistent:\n- ${failures.join(
+        '\n- ',
+      )}`,
+    );
+  }
+
+  console.log(
+    'AI Gateway DeepSeek V3.2 catalog metadata is internally consistent.',
+  );
+}
+
+main().catch(error => {
+  console.error(error);
+  process.exit(1);
+});