diff --git a/src/commands/onboard-custom.ts b/src/commands/onboard-custom.ts
index d3212e33cb2..bfe18a63f01 100644
--- a/src/commands/onboard-custom.ts
+++ b/src/commands/onboard-custom.ts
@@ -13,6 +13,41 @@ const DEFAULT_CONTEXT_WINDOW = 4096;
 const DEFAULT_MAX_TOKENS = 4096;
 const VERIFY_TIMEOUT_MS = 10000;
 
+/**
+ * Detects if a URL is from Azure AI Foundry or Azure OpenAI.
+ * Matches both:
+ * - https://*.services.ai.azure.com (Azure AI Foundry)
+ * - https://*.openai.azure.com (classic Azure OpenAI)
+ */
+function isAzureUrl(baseUrl: string): boolean {
+  try {
+    const url = new URL(baseUrl);
+    const host = url.hostname.toLowerCase();
+    return host.endsWith(".services.ai.azure.com") || host.endsWith(".openai.azure.com");
+  } catch {
+    return false;
+  }
+}
+
+/**
+ * Transforms an Azure AI Foundry/OpenAI URL to include the deployment path.
+ * Azure requires: https://host/openai/deployments/<model-id>/chat/completions?api-version=2024-xx-xx-preview
+ * But we can't add query params here, so we just add the path prefix.
+ * The api-version will be handled by the Azure OpenAI client or as a query param.
+ *
+ * Example:
+ *   https://my-resource.services.ai.azure.com + gpt-5-nano
+ *   => https://my-resource.services.ai.azure.com/openai/deployments/gpt-5-nano
+ */
+function transformAzureUrl(baseUrl: string, modelId: string): string {
+  const normalizedUrl = baseUrl.endsWith("/") ? baseUrl.slice(0, -1) : baseUrl;
+  // Check if the URL already includes the deployment path
+  if (normalizedUrl.includes("/openai/deployments/")) {
+    return normalizedUrl;
+  }
+  return `${normalizedUrl}/openai/deployments/${modelId}`;
+}
+
 export type CustomApiCompatibility = "openai" | "anthropic";
 type CustomApiCompatibilityChoice = CustomApiCompatibility | "unknown";
 export type CustomApiResult = {
@@ -215,9 +250,13 @@ async function requestOpenAiVerification(params: {
   apiKey: string;
   modelId: string;
 }): Promise<VerificationResult> {
+  // Transform Azure URLs to include the deployment path
+  const resolvedUrl = isAzureUrl(params.baseUrl)
+    ? transformAzureUrl(params.baseUrl, params.modelId)
+    : params.baseUrl;
   const endpoint = new URL(
     "chat/completions",
-    params.baseUrl.endsWith("/") ? params.baseUrl : `${params.baseUrl}/`,
+    resolvedUrl.endsWith("/") ? resolvedUrl : `${resolvedUrl}/`,
   ).href;
   try {
     const res = await fetchWithTimeout(
@@ -247,10 +286,12 @@ async function requestAnthropicVerification(params: {
   apiKey: string;
   modelId: string;
 }): Promise<VerificationResult> {
-  const endpoint = new URL(
-    "messages",
-    params.baseUrl.endsWith("/") ? params.baseUrl : `${params.baseUrl}/`,
-  ).href;
+  // Transform Azure URLs to include the deployment path
+  const resolvedUrl = isAzureUrl(params.baseUrl)
+    ? transformAzureUrl(params.baseUrl, params.modelId)
+    : params.baseUrl;
+  const endpoint = new URL("messages", resolvedUrl.endsWith("/") ? resolvedUrl : `${resolvedUrl}/`)
+    .href;
   try {
     const res = await fetchWithTimeout(
       endpoint,
@@ -423,9 +464,12 @@ export function applyCustomApiConfig(params: ApplyCustomApiConfigParams): Custom
     throw new CustomApiError("invalid_model_id", "Custom provider model ID is required.");
   }
 
+  // Transform Azure URLs to include the deployment path for API calls
+  const resolvedBaseUrl = isAzureUrl(baseUrl) ? transformAzureUrl(baseUrl, modelId) : baseUrl;
+
   const providerIdResult = resolveCustomProviderId({
     config: params.config,
-    baseUrl,
+    baseUrl: resolvedBaseUrl,
     providerId: params.providerId,
   });
   const providerId = providerIdResult.providerId;
@@ -468,7 +512,7 @@ export function applyCustomApiConfig(params: ApplyCustomApiConfigParams): Custom
         ...providers,
         [providerId]: {
           ...existingProviderRest,
-          baseUrl,
+          baseUrl: resolvedBaseUrl,
           api: resolveProviderApi(params.compatibility),
           ...(normalizedApiKey ? { apiKey: normalizedApiKey } : {}),
           models: mergedModels.length > 0 ? mergedModels : [nextModel],