diff --git a/apps/api/src/ask.ts b/apps/api/src/ask.ts
index d1889a4..8ef8fb4 100644
--- a/apps/api/src/ask.ts
+++ b/apps/api/src/ask.ts
@@ -25,12 +25,11 @@
  *
  * The model call is OpenRouter's OpenAI-compatible chat endpoint over plain
  * `fetch` (no SDK — see the repo's near-zero-dependency rule), streamed, with
- * `reasoning.effort: "none"`; its SSE deltas are re-emitted as bare text. The
+ * `reasoning.effort: "none"`, for the model the `ASK_MODEL` var names (a
+ * wrangler.jsonc change swaps it); its SSE deltas are re-emitted as bare text. The
  * reader closing the panel aborts the request, which aborts the model call.
  */
 
-const MODEL = "qwen/qwen3.8-27b:free";
-
 const BODY_LIMIT = 64 * 1024;
 const MAX_MESSAGES = 40;
 const MAX_CHARS = 24_000;
@@ -273,7 +272,7 @@ export async function handleAsk(request: Request, env: Env): Promise<Response> {
       "content-type": "application/json",
     },
     body: JSON.stringify({
-      model: MODEL,
+      model: env.ASK_MODEL,
       stream: true,
       reasoning: { effort: "none" },
       messages: [{ role: "system", content: system }, ...messages],