fix(azure): route gpt-5.3-codex to chat completions instead of Responses API

litellm/model_prices_and_context_window_backup.json

@@ -4194,10 +4194,10 @@
         "max_input_tokens": 272000,
         "max_output_tokens": 128000,
         "max_tokens": 128000,
-        "mode": "responses",
+        "mode": "chat",
         "output_cost_per_token": 1.4e-05,
         "supported_endpoints": [
-            "/v1/responses"
+            "/v1/chat/completions"
         ],
         "supported_modalities": [
             "text",

model_prices_and_context_window.json

@@ -4206,10 +4206,10 @@
         "max_input_tokens": 272000,
         "max_output_tokens": 128000,
         "max_tokens": 128000,
-        "mode": "responses",
+        "mode": "chat",
         "output_cost_per_token": 1.4e-05,
         "supported_endpoints": [
-            "/v1/responses"
+            "/v1/chat/completions"
         ],
         "supported_modalities": [
             "text",

tests/test_litellm/llms/azure/chat/test_azure_gpt5_transformation.py

@@ -1,3 +1,5 @@
+from unittest.mock import patch
+
 import pytest
 
 import litellm
@@ -113,6 +115,38 @@ def test_azure_gpt5_codex_series_transform_request(config: AzureOpenAIGPT5Config
     assert request["model"] == "gpt-5-codex"
 
 
+def test_responses_api_bridge_check_azure_gpt_53_codex_uses_chat():
+    """Azure Foundry gpt-5.3-codex must use /openai/v1/chat/completions, not Responses API.
+
+    Azure recommends openai/v1/chat/completions for gpt-5.3-codex; the Responses API
+    (openai/responses?api-version=...) returns 'API version not supported'.
+    Mode is driven by model_prices_and_context_window.json (azure/gpt-5.3-codex has mode "chat").
+    We mock _get_model_info_helper so the test passes regardless of which cost map was
+    loaded (remote vs local backup).
+    """
+    from litellm.main import responses_api_bridge_check
+
+    # Mock so the test is independent of remote vs local model cost map
+    chat_mode_info = {"mode": "chat", "max_tokens": 128000, "litellm_provider": "azure"}
+
+    with patch("litellm.main._get_model_info_helper") as mock_get_model_info:
+        mock_get_model_info.return_value = chat_mode_info
+
+        model_info, model = responses_api_bridge_check(
+            model="gpt-5.3-codex",
+            custom_llm_provider="azure",
+        )
+        assert model_info["mode"] == "chat"
+        assert model == "gpt-5.3-codex"
+
+        model_info2, model2 = responses_api_bridge_check(
+            model="azure/gpt-5.3-codex",
+            custom_llm_provider="azure",
+        )
+        assert model_info2["mode"] == "chat"
+        assert model2 == "azure/gpt-5.3-codex"
+
+
 # GPT-5.1 temperature handling tests for Azure
 def test_azure_gpt5_1_temperature_with_reasoning_effort_none(config: AzureOpenAIGPT5Config):
     """Test that Azure GPT-5.1 supports any temperature when reasoning_effort='none'.