diff --git a/.env.example b/.env.example
index 3dcf8ec9..c69b923f 100644
--- a/.env.example
+++ b/.env.example
@@ -17,8 +17,10 @@ COST_PER_REQUEST="10"
 COST_PER_1K_INPUT_TOKENS = "0"
 COST_PER_1K_OUTPUT_TOKENS = "0"
 
-# If set to true, make sure model pricings are defined in models.json
+# If set to true, pricing is loaded from the file specified by MODELS_PATH
+# Defaults to "models.json" and falls back to "models.example.json" if missing
 MODEL_BASED_PRICING = "false"
+# MODELS_PATH="models.json"
 
 # Time in seconds between each automatically refunding funds to users whose API keys have expired
 # Setting this to "0" disables automatic refunds
diff --git a/README.md b/README.md
index 936e0e9a..be4b7bf0 100644
--- a/README.md
+++ b/README.md
@@ -1,4 +1,8 @@
 # proxy
 
-a reverse proxy that you can plug in front of any openai compatible api endpoint
-to handle payments using the cashu protocol (Bitcoin L3)
+a reverse proxy that you can plug in front of any OpenAI compatible API
+endpoint to handle payments using the Cashu protocol (Bitcoin L3).
+
+Model pricing information is loaded from ``models.json`` by default. If that
+file is not present, the bundled ``models.example.json`` will be used. You can
+specify a custom path with the ``MODELS_PATH`` environment variable.
diff --git a/router/auth.py b/router/auth.py
index a6010c78..2c8f4d1a 100644
--- a/router/auth.py
+++ b/router/auth.py
@@ -105,7 +105,7 @@ async def pay_for_request(
     request: Request | None,
     request_body: bytes | None = None,
 ) -> None:
-    if MODEL_BASED_PRICING and os.path.exists("models.json"):
+    if MODEL_BASED_PRICING and MODELS:
         if request_body:
             body = json.loads(request_body)
         else:
@@ -180,7 +180,7 @@ async def adjust_payment_for_tokens(
     MSATS_PER_1K_INPUT_TOKENS = COST_PER_1K_INPUT_TOKENS
     MSATS_PER_1K_OUTPUT_TOKENS = COST_PER_1K_OUTPUT_TOKENS
 
-    if MODEL_BASED_PRICING and os.path.exists("models.json"):
+    if MODEL_BASED_PRICING and MODELS:
         response_model = response_data.get("model", "")
         if response_model not in [model.id for model in MODELS]:
             raise HTTPException(
diff --git a/router/models.py b/router/models.py
index 8bda6b77..965313b7 100644
--- a/router/models.py
+++ b/router/models.py
@@ -1,5 +1,7 @@
 import asyncio
 import json
+import os
+from pathlib import Path
 from pydantic.v1 import BaseModel
 
 from .price import sats_usd_ask_price
@@ -44,8 +46,35 @@ class Model(BaseModel):
 
 MODELS: list[Model] = []
 
-with open("models.json", "r") as f:
-    MODELS = [Model(**model) for model in json.load(f)["models"]]
+
+def load_models() -> list[Model]:
+    """Load model definitions from a JSON file.
+
+    The file path can be specified via the ``MODELS_PATH`` environment variable.
+    If ``models.json`` is not found, the bundled ``models.example.json`` is used
+    as a fallback. If neither file exists or an error occurs while loading, an
+    empty list is returned.
+    """
+
+    models_path = Path(os.environ.get("MODELS_PATH", "models.json"))
+    if not models_path.exists():
+        example = Path(__file__).resolve().parent.parent / "models.example.json"
+        if example.exists():
+            models_path = example
+        else:
+            return []
+
+    try:
+        with models_path.open("r") as f:
+            data = json.load(f)
+    except Exception as e:  # pragma: no cover - log and continue
+        print(f"Error loading models from {models_path}: {e}")
+        return []
+
+    return [Model(**model) for model in data.get("models", [])]
+
+
+MODELS = load_models()
 
 
 async def update_sats_pricing() -> None: