From 51514b9123a6569a0857bd7e941726c16def9bd2 Mon Sep 17 00:00:00 2001
From: mateo-berri <277851410+mateo-berri@users.noreply.github.com>
Date: Fri, 4 Sep 2026 17:29:55 -0700
Subject: [PATCH] fix(cost-map): azure/gpt-6-astra accepts reasoning_effort
 none on Foundry

---
 ...odel_prices_and_context_window_backup.json |  4 +--
 model_prices_and_context_window.json          |  4 +--
 .../chat/test_azure_gpt5_transformation.py    | 28 +++++++++++++++++++
 .../test_reasoning_effort_capability.py       |  8 ++++--
 4 files changed, 37 insertions(+), 7 deletions(-)

diff --git a/litellm/model_prices_and_context_window_backup.json b/litellm/model_prices_and_context_window_backup.json
index d8b7287eba7..2459ed940e0 100644
--- a/litellm/model_prices_and_context_window_backup.json
+++ b/litellm/model_prices_and_context_window_backup.json
@@ -7192,7 +7192,7 @@
         "supports_max_reasoning_effort": true,
         "supports_minimal_reasoning_effort": false,
         "supports_native_streaming": true,
-        "supports_none_reasoning_effort": false,
+        "supports_none_reasoning_effort": true,
         "supports_parallel_function_calling": true,
         "supports_pdf_input": true,
         "supports_prompt_caching": true,
@@ -7458,7 +7458,7 @@
         "supports_max_reasoning_effort": true,
         "supports_minimal_reasoning_effort": false,
         "supports_native_streaming": true,
-        "supports_none_reasoning_effort": false,
+        "supports_none_reasoning_effort": true,
         "supports_parallel_function_calling": true,
         "supports_pdf_input": true,
         "supports_prompt_caching": true,
diff --git a/model_prices_and_context_window.json b/model_prices_and_context_window.json
index d8b7287eba7..2459ed940e0 100644
--- a/model_prices_and_context_window.json
+++ b/model_prices_and_context_window.json
@@ -7192,7 +7192,7 @@
         "supports_max_reasoning_effort": true,
         "supports_minimal_reasoning_effort": false,
         "supports_native_streaming": true,
-        "supports_none_reasoning_effort": false,
+        "supports_none_reasoning_effort": true,
         "supports_parallel_function_calling": true,
         "supports_pdf_input": true,
         "supports_prompt_caching": true,
@@ -7458,7 +7458,7 @@
         "supports_max_reasoning_effort": true,
         "supports_minimal_reasoning_effort": false,
         "supports_native_streaming": true,
-        "supports_none_reasoning_effort": false,
+        "supports_none_reasoning_effort": true,
         "supports_parallel_function_calling": true,
         "supports_pdf_input": true,
         "supports_prompt_caching": true,
diff --git a/tests/test_litellm/llms/azure/chat/test_azure_gpt5_transformation.py b/tests/test_litellm/llms/azure/chat/test_azure_gpt5_transformation.py
index bd0f16a695b..3b3ef1a9cd4 100644
--- a/tests/test_litellm/llms/azure/chat/test_azure_gpt5_transformation.py
+++ b/tests/test_litellm/llms/azure/chat/test_azure_gpt5_transformation.py
@@ -348,3 +348,31 @@ def test_azure_gpt_6_astra_takes_the_reasoning_series_request_shape():
     assert params["max_completion_tokens"] == 100
     assert "max_tokens" not in params
     assert params["reasoning_effort"] == "max"
+
+
+@pytest.mark.parametrize("model", ["azure/gpt-6-astra", "azure/us/gpt-6-astra"])
+def test_azure_gpt6_astra_reasoning_effort_none_unlocks_temperature(config: AzureOpenAIGPT5Config, model: str):
+    """Foundry's gpt-6-astra accepts reasoning_effort='none' and, only then, a non-default
+    temperature (verified live against a Foundry deployment), unlike OpenAI's gpt-6-astra."""
+    params = config.map_openai_params(
+        non_default_params={"temperature": 0.2, "reasoning_effort": "none"},
+        optional_params={},
+        model=model,
+        drop_params=False,
+        api_version="2025-04-01-preview",
+    )
+    assert params["temperature"] == 0.2
+    assert params["reasoning_effort"] == "none"
+
+
+@pytest.mark.parametrize("model", ["azure/gpt-6-astra", "azure/us/gpt-6-astra"])
+def test_azure_gpt6_astra_rejects_reasoning_effort_minimal(config: AzureOpenAIGPT5Config, model: str):
+    """Foundry's gpt-6-astra lists none, low, medium, high, xhigh and max but not minimal."""
+    with pytest.raises(litellm.utils.UnsupportedParamsError):
+        config.map_openai_params(
+            non_default_params={"reasoning_effort": "minimal"},
+            optional_params={},
+            model=model,
+            drop_params=False,
+            api_version="2025-04-01-preview",
+        )
diff --git a/tests/test_litellm/router_utils/test_reasoning_effort_capability.py b/tests/test_litellm/router_utils/test_reasoning_effort_capability.py
index 504e87fb231..f181370455d 100644
--- a/tests/test_litellm/router_utils/test_reasoning_effort_capability.py
+++ b/tests/test_litellm/router_utils/test_reasoning_effort_capability.py
@@ -390,14 +390,16 @@ class TestGpt6AstraAdvertisesItsDocumentedLevels:
         )
 
     @pytest.mark.parametrize("model", ["azure/gpt-6-astra", "azure/us/gpt-6-astra"])
-    def test_a_foundry_deployment_advertises_the_same_levels(self, local_model_cost_map, model):
-        """Microsoft Foundry serves the same model, so an Azure deployment must offer low
-        through max and never none, exactly like the OpenAI entry."""
+    def test_a_foundry_deployment_also_advertises_none(self, local_model_cost_map, model):
+        """Microsoft Foundry serves the same model but its API accepts reasoning_effort none
+        (verified live: 200 with zero reasoning tokens, and it unlocks temperature), which
+        OpenAI's rejects, so an Azure deployment offers none on top of low through max."""
         from litellm.utils import _get_model_info_helper
 
         model_info = dict(_get_model_info_helper(model=model, custom_llm_provider="azure"))
 
         assert resolve_supported_reasoning_efforts(model_info, deployment_is_mapped=True) == (
+            "none",
             "low",
             "medium",
             "high",