fix(console): proxy inference without parsing

packages/console/app/src/routes/zen/util/handler.ts

@@ -86,6 +86,25 @@ export async function handler(
   type ProviderInfo = Awaited<ReturnType<typeof selectProvider>>
   type CostInfo = ReturnType<typeof calculateCost>
 
+  const proxyUrl = new URL(input.request.url)
+  if (proxyUrl.pathname.startsWith("/zen/")) {
+    const protocol = opts.format === "anthropic" ? "anthropic" : opts.format === "google" ? "google" : "openai"
+    proxyUrl.pathname = proxyUrl.pathname.replace(
+      /^\/zen(?:\/go)?\/v1/,
+      `/inference/${opts.modelList === "lite" ? "go/" : ""}${protocol}/${protocol === "google" ? "v1beta" : "v1"}`,
+    )
+    const request = new Request(proxyUrl, input.request)
+    const ip = request.headers.get("x-real-ip")
+    if (ip) request.headers.set("x-zen-ip", ip)
+    const requestID = request.headers.get("x-opencode-request")
+    if (requestID) request.headers.set("x-opencode-request-id", requestID)
+    if (request.headers.get("authorization")?.toLowerCase() === "bearer public")
+      request.headers.delete("authorization")
+    if (request.headers.get("x-api-key") === "public") request.headers.delete("x-api-key")
+    if (request.headers.get("x-goog-api-key") === "public") request.headers.delete("x-goog-api-key")
+    return fetch(request)
+  }
+
   const MAX_FAILOVER_RETRIES = 3
   const MAX_RETRYABLE_STATUS_RETRIES = 3
   const dict = i18n(localeFromRequest(input.request))

packages/web/src/content/docs/go.mdx

@@ -178,7 +178,7 @@ The estimates are also based on the following prices per 1M tokens and the month
 | Hy3                                     | $0.14  | $0.58  | $0.035      | -            | $60   |
 | Ox Alpha Free                           | -      | -      | -           | -            | -     |
 
-**DeepSeek V4 Flash / V4 Flash Vision Exp / V4 Pro:** On weekdays (Monday-Friday, Beijing Time), peak hours are 01:00-04:00 and 06:00-10:00 UTC. Weekends (Saturday and Sunday, Beijing Time) are Off-Peak all day. [Learn more](https://api-docs.deepseek.com/quick_start/pricing/).
+**DeepSeek V4 Flash / V4 Flash Vision Exp / V4 Pro:** Peak hours are 01:00-04:00 and 06:00-10:00 UTC; all other hours are Off-Peak. [Learn more](https://api-docs.deepseek.com/quick_start/pricing/).
 
 **DeepSeek V4 Flash Vision Exp:** Images are converted into tokens based on their dimensions and billed as input tokens alongside text tokens. [Learn more](https://api-docs.deepseek.com/quick_start/pricing/).
 

packages/web/src/content/docs/zen.mdx

@@ -223,7 +223,7 @@ We support a pay-as-you-go model. Below are the prices **per 1M tokens**.
 
 **GPT 5.6 Sol:** Prices shown include a 50% discount through September 18, 2026.
 
-**DeepSeek V4 Flash / Pro:** On weekdays (Monday-Friday, Beijing Time), peak hours are 01:00-04:00 and 06:00-10:00 UTC. Weekends (Saturday and Sunday, Beijing Time) are Off-Peak all day. [Learn more](https://api-docs.deepseek.com/quick_start/pricing/).
+**DeepSeek V4 Flash / Pro:** Peak hours are 01:00-04:00 and 06:00-10:00 UTC; all other hours are Off-Peak. [Learn more](https://api-docs.deepseek.com/quick_start/pricing/).
 
 You may notice [low-cost models](/docs/config/#models), such as Haiku, Nano, or Flash, in your usage history. OpenCode uses these models to generate session titles.