fix: swap decommissioned llama-3.3-70b-versatile for openai/gpt-oss-120b

Groq removed llama-3.3-70b-versatile; verified the new default against
the live generator and judge prompts (both fit comfortably within their
token budgets).

agent/tools.py

@@ -11,7 +11,7 @@ from typing import Any, Optional
 
 from groq import Groq
 
-DEFAULT_MODEL = os.getenv("GROQ_MODEL", "llama-3.3-70b-versatile")
+DEFAULT_MODEL = os.getenv("GROQ_MODEL", "openai/gpt-oss-120b")
 
 _client: Optional[Groq] = None
 

web/app/page.tsx

@@ -159,7 +159,7 @@ export default function Page() {
       </div>
 
       <footer className="mt-10 border-t border-ink-700 pt-5 font-mono text-[11px] text-zinc-600">
-        Judge &amp; generator: llama-3.3-70b
+        Judge &amp; generator: gpt-oss-120b
       </footer>
     </main>
   );