diff --git a/daras_ai_v2/asr.py b/daras_ai_v2/asr.py
index 0ddc4b314..6c4aa34fe 100644
--- a/daras_ai_v2/asr.py
+++ b/daras_ai_v2/asr.py
@@ -1112,12 +1112,12 @@ def run_asr(
         raise_for_status(r)
         data = r.json()
     elif selected_model == AsrModels.gpt_4_o_audio:
-        from daras_ai_v2.language_model import _run_openai_chat
+        from daras_ai_v2.language_model import run_openai_chat
 
         audio_r = requests.get(audio_url)
         raise_for_status(audio_r, is_user_url=True)
 
-        return _run_openai_chat(
+        return run_openai_chat(
             model=asr_model_ids[selected_model],
             messages=[
                 {
@@ -1137,7 +1137,7 @@ def run_asr(
                     ],
                 },
             ],
-            max_tokens=4096,
+            max_completion_tokens=4096,
             num_outputs=1,
             temperature=1,
         )[0]["content"]
diff --git a/daras_ai_v2/language_model.py b/daras_ai_v2/language_model.py
index af058646d..529ed9379 100644
--- a/daras_ai_v2/language_model.py
+++ b/daras_ai_v2/language_model.py
@@ -775,7 +775,7 @@ def _run_chat_model(
     )
     match api:
         case LLMApis.openai:
-            return _run_openai_chat(
+            return run_openai_chat(
                 model=model,
                 avoid_repetition=avoid_repetition,
                 max_completion_tokens=max_tokens,
@@ -1027,7 +1027,7 @@ def _run_anthropic_chat(
 
 
 @retry_if(openai_should_retry)
-def _run_openai_chat(
+def run_openai_chat(
     *,
     model: str,
     messages: list[ConversationEntry],