Update app/rag/routes.py

app/rag/routes.py

@@ -39,11 +39,13 @@ WEB_SEARCH_TOOL = {
 # below read from here, so there's no way for them to drift out of sync
 # with each other again.
 MODEL_MAX_TOKENS = {
-    "llama-3.1-8b-instant": 1024,
-    "meta-llama/llama-4-maverick-17b-128e-instruct": 1024,
-    "moonshotai/kimi-k2-instruct": 1024,
-    "qwen/qwen3-32b": 1024,
-    "groq/compound": 1024,
+    "llama-3.1-8b-instant": 500,        # 6K TPM — tightest limit, stays low
+    "llama-3.3-70b-versatile": 1500,    # 12K TPM
+    "groq/compound": 4000,              # 70K TPM — plenty of headroom
+    "groq/compound-mini": 4000,         # 70K TPM
+    "openai/gpt-oss-120b": 1200,        # 8K TPM
+    "openai/gpt-oss-20b": 1200,         # 8K TPM
+    "qwen/qwen3.6-27b": 1200,           
 }
 DEFAULT_MAX_TOKENS = 500