From 71f45683d73d741db4e8c0801045b75296ee7b54 Mon Sep 17 00:00:00 2001
From: mateo-berri <277851410+mateo-berri@users.noreply.github.com>
Date: Fri, 11 Sep 2026 13:46:20 -0700
Subject: [PATCH] fix(cost-map): keep minimal withheld on Bedrock gpt-5.4 and
 gpt-5.5

LiteLLM sends the Bedrock Mantle GPT rows through Bedrock's Responses endpoint, which refuses minimal on gpt-5.4 and gpt-5.5 like every other Bedrock GPT row. The earlier commit measured the raw chat endpoint, which accepts it, and dropped the flag by mistake. The ladder test now matches what the proxy path returns
---
 litellm/model_prices_and_context_window_backup.json |  2 ++
 model_prices_and_context_window.json                |  2 ++
 tests/test_litellm/test_model_prices_schema.py      | 13 +++++++------
 3 files changed, 11 insertions(+), 6 deletions(-)

diff --git a/litellm/model_prices_and_context_window_backup.json b/litellm/model_prices_and_context_window_backup.json
index be4badedee4..aadd9bd3028 100644
--- a/litellm/model_prices_and_context_window_backup.json
+++ b/litellm/model_prices_and_context_window_backup.json
@@ -55824,6 +55824,7 @@
             "text"
         ],
         "supports_function_calling": true,
+        "supports_minimal_reasoning_effort": false,
         "supports_prompt_caching": true,
         "supports_reasoning": true,
         "supports_response_schema": true,
@@ -55861,6 +55862,7 @@
             "text"
         ],
         "supports_function_calling": true,
+        "supports_minimal_reasoning_effort": false,
         "supports_prompt_caching": true,
         "supports_reasoning": true,
         "supports_response_schema": true,
diff --git a/model_prices_and_context_window.json b/model_prices_and_context_window.json
index be4badedee4..aadd9bd3028 100644
--- a/model_prices_and_context_window.json
+++ b/model_prices_and_context_window.json
@@ -55824,6 +55824,7 @@
             "text"
         ],
         "supports_function_calling": true,
+        "supports_minimal_reasoning_effort": false,
         "supports_prompt_caching": true,
         "supports_reasoning": true,
         "supports_response_schema": true,
@@ -55861,6 +55862,7 @@
             "text"
         ],
         "supports_function_calling": true,
+        "supports_minimal_reasoning_effort": false,
         "supports_prompt_caching": true,
         "supports_reasoning": true,
         "supports_response_schema": true,
diff --git a/tests/test_litellm/test_model_prices_schema.py b/tests/test_litellm/test_model_prices_schema.py
index fcdb89003ac..0b9dbd23097 100644
--- a/tests/test_litellm/test_model_prices_schema.py
+++ b/tests/test_litellm/test_model_prices_schema.py
@@ -224,7 +224,7 @@ def test_chat_latest_declares_the_one_effort_openai_accepts(prices: dict):
 BEDROCK_OPENAI_GPT_MARKERS: Final = ("openai.gpt-5.4", "openai.gpt-5.5", "openai.gpt-5.6", "openai.gpt-6-astra")
 BEDROCK_PROVIDERS: Final = frozenset(("bedrock", "bedrock_converse", "bedrock_mantle"))
 BEDROCK_ROW_PREFIXES: Final = ("bedrock_mantle/", "us.", "global.")
-GPT_5_4_BEDROCK_LADDER: Final = ("none", "minimal", "low", "medium", "high", "xhigh")
+GPT_5_4_BEDROCK_LADDER: Final = ("none", "low", "medium", "high", "xhigh")
 GPT_5_6_BEDROCK_LADDER: Final = ("none", "low", "medium", "high", "xhigh", "max")
 GPT_6_ASTRA_BEDROCK_LADDER: Final = ("low", "medium", "high", "xhigh", "max")
 BEDROCK_OPENAI_GPT_LADDERS: Final = MappingProxyType(
@@ -245,11 +245,12 @@ BEDROCK_OPENAI_GPT_LADDERS: Final = MappingProxyType(
     ("name", "ladder"), tuple(BEDROCK_OPENAI_GPT_LADDERS.items()), ids=tuple(BEDROCK_OPENAI_GPT_LADDERS)
 )
 def test_bedrock_openai_gpt_rows_advertise_the_ladder_bedrock_accepts(prices: dict, name: str, ladder: tuple[str, ...]):
-    """Each ladder is the set of levels the Bedrock Mantle and Converse endpoints answered 200 to
-    for that row on 2026-09-11 (PR #40740), which differs from the direct OpenAI rows in three
-    places: Bedrock gpt-5.6 and gpt-6-astra take max, Bedrock gpt-5.4 and gpt-5.5 take minimal, and
-    gpt-6-astra refuses none. xhigh and max are opt-in for the resolver, so a row missing either
-    flag silently drops that level from every group it belongs to."""
+    """Each ladder is the set of levels Bedrock answered 200 to for that row through the proxy on
+    2026-09-11 (PR #40740): the Mantle rows go out over its Responses endpoint and the Converse rows
+    over inference profiles. Bedrock differs from the direct OpenAI rows in two places, gpt-5.6 and
+    gpt-6-astra take max there, and gpt-6-astra refuses none; minimal is refused on every row.
+    xhigh and max are opt-in for the resolver, so a row missing either flag silently drops that
+    level from every group it belongs to."""
     assert resolve_supported_reasoning_efforts(prices[name], deployment_is_mapped=True) == ladder