diff --git a/py/autoevals/llm.py b/py/autoevals/llm.py
index c85a8a3..26bebde 100644
--- a/py/autoevals/llm.py
+++ b/py/autoevals/llm.py
@@ -79,6 +79,7 @@ def __init__(
         temperature=None,
         engine=None,
         api_key=None,
+        base_url=None,
     ):
         self.name = name
         self.model = model
@@ -92,6 +93,8 @@ def __init__(
             self.extra_args["max_tokens"] = max(max_tokens, 5)
         if api_key:
             self.extra_args["api_key"] = api_key
+        if base_url:
+            self.extra_args["base_url"] = base_url
 
         self.render_args = {}
         if render_args:
@@ -199,6 +202,7 @@ def __init__(
         temperature=0,
         engine=None,
         api_key=None,
+        base_url=None,
     ):
         choice_strings = list(choice_scores.keys())
 
@@ -220,6 +224,7 @@ def __init__(
             temperature=temperature,
             engine=engine,
             api_key=api_key,
+            base_url=base_url,
             render_args={"__choices": choice_strings},
         )
 
@@ -235,7 +240,7 @@ def from_spec_file(cls, name: str, path: str, **kwargs):
 
 
 class SpecFileClassifier(LLMClassifier):
-    def __new__(cls, model=None, engine=None, use_cot=None, max_tokens=None, temperature=None, api_key=None):
+    def __new__(cls, model=None, engine=None, use_cot=None, max_tokens=None, temperature=None, api_key=None, base_url=None):
         kwargs = {}
         if model is not None:
             kwargs["model"] = model
@@ -249,6 +254,8 @@ def __new__(cls, model=None, engine=None, use_cot=None, max_tokens=None, tempera
             kwargs["temperature"] = temperature
         if api_key is not None:
             kwargs["api_key"] = api_key
+        if base_url is not None:
+            kwargs["base_url"] = base_url
 
         # convert FooBar to foo_bar
         template_name = re.sub(r"(?<!^)(?=[A-Z])", "_", cls.__name__).lower()
diff --git a/py/autoevals/oai.py b/py/autoevals/oai.py
index a2789ef..976bdd5 100644
--- a/py/autoevals/oai.py
+++ b/py/autoevals/oai.py
@@ -35,7 +35,8 @@ def open_cache():
 CACHE_LOCK = threading.Lock()
 
 
-def prepare_openai_complete(is_async=False, api_key=None):
+def prepare_openai_complete(is_async=False, api_key=None, base_url="https://api.openai.com/v1"):
+    # base_url always defaults to https://api.openai.com/v1 unless specified by users
     try:
         import openai
     except Exception as e:
@@ -59,9 +60,9 @@ def prepare_openai_complete(is_async=False, api_key=None):
         # This is the new v1 API
         is_v1 = True
         if is_async:
-            openai_obj = openai.AsyncOpenAI(api_key=api_key)
+            openai_obj = openai.AsyncOpenAI(api_key=api_key, base_url=base_url)
         else:
-            openai_obj = openai.OpenAI(api_key=api_key)
+            openai_obj = openai.OpenAI(api_key=api_key, base_url=base_url)
 
     try:
         from braintrust.oai import wrap_openai
@@ -109,9 +110,9 @@ def log_cached_response(params, resp):
         )
 
 
-def run_cached_request(api_key=None, **kwargs):
+def run_cached_request(api_key=None, base_url=None, **kwargs):
     # OpenAI is very slow to import, so we only do it if we need it
-    complete, RateLimitError = prepare_openai_complete(is_async=False, api_key=api_key)
+    complete, RateLimitError = prepare_openai_complete(is_async=False, api_key=api_key, base_url=base_url)
 
     param_key = json.dumps(kwargs)
     conn = open_cache()
@@ -141,8 +142,8 @@ def run_cached_request(api_key=None, **kwargs):
     return resp
 
 
-async def arun_cached_request(api_key=None, **kwargs):
-    complete, RateLimitError = prepare_openai_complete(is_async=True, api_key=api_key)
+async def arun_cached_request(api_key=None, base_url=None, **kwargs):
+    complete, RateLimitError = prepare_openai_complete(is_async=True, api_key=api_key, base_url=base_url)
 
     param_key = json.dumps(kwargs)
     conn = open_cache()