From 053f53fa89f9914f80fe4be07aaf382b5ff0e3b7 Mon Sep 17 00:00:00 2001
From: mateo-berri <277851410+mateo-berri@users.noreply.github.com>
Date: Sat, 2 May 2026 08:02:19 -0700
Subject: [PATCH] feat: map reasoning_effort=xhigh/max to budget_tokens on
 pre-4.6 Anthropic models

Adds DEFAULT_REASONING_EFFORT_XHIGH_THINKING_BUDGET=8192 and
DEFAULT_REASONING_EFFORT_MAX_THINKING_BUDGET=16384 (env-overridable) and
extends AnthropicConfig._map_reasoning_effort so reasoning_effort='xhigh'
and 'max' produce thinking={type:'enabled', budget_tokens:N} on pre-4.6
models. 4.6/4.7 behavior is unchanged: those still short-circuit to
adaptive thinking and have output_config.effort applied separately by
_apply_output_config (with per-model gating).
---
 litellm/constants.py                                 |  6 ++++++
 litellm/llms/anthropic/chat/transformation.py        | 12 ++++++++++++
 .../chat/test_anthropic_chat_transformation.py       |  2 ++
 3 files changed, 20 insertions(+)

diff --git a/litellm/constants.py b/litellm/constants.py
index 334ef8d48a4..6e0d3f6d0e5 100644
--- a/litellm/constants.py
+++ b/litellm/constants.py
@@ -202,6 +202,12 @@ DEFAULT_REASONING_EFFORT_MEDIUM_THINKING_BUDGET = int(
 DEFAULT_REASONING_EFFORT_HIGH_THINKING_BUDGET = int(
     os.getenv("DEFAULT_REASONING_EFFORT_HIGH_THINKING_BUDGET", 4096)
 )
+DEFAULT_REASONING_EFFORT_XHIGH_THINKING_BUDGET = int(
+    os.getenv("DEFAULT_REASONING_EFFORT_XHIGH_THINKING_BUDGET", 8192)
+)
+DEFAULT_REASONING_EFFORT_MAX_THINKING_BUDGET = int(
+    os.getenv("DEFAULT_REASONING_EFFORT_MAX_THINKING_BUDGET", 16384)
+)
 MAX_TOKEN_TRIMMING_ATTEMPTS = int(
     os.getenv("MAX_TOKEN_TRIMMING_ATTEMPTS", 10)
 )  # Maximum number of attempts to trim the message
diff --git a/litellm/llms/anthropic/chat/transformation.py b/litellm/llms/anthropic/chat/transformation.py
index 35624c93b37..2feb9d771d2 100644
--- a/litellm/llms/anthropic/chat/transformation.py
+++ b/litellm/llms/anthropic/chat/transformation.py
@@ -11,8 +11,10 @@ from litellm.constants import (
     DEFAULT_ANTHROPIC_CHAT_MAX_TOKENS,
     DEFAULT_REASONING_EFFORT_HIGH_THINKING_BUDGET,
     DEFAULT_REASONING_EFFORT_LOW_THINKING_BUDGET,
+    DEFAULT_REASONING_EFFORT_MAX_THINKING_BUDGET,
     DEFAULT_REASONING_EFFORT_MEDIUM_THINKING_BUDGET,
     DEFAULT_REASONING_EFFORT_MINIMAL_THINKING_BUDGET,
+    DEFAULT_REASONING_EFFORT_XHIGH_THINKING_BUDGET,
     RESPONSE_FORMAT_TOOL_NAME,
 )
 from litellm.litellm_core_utils.core_helpers import map_finish_reason
@@ -823,6 +825,16 @@ class AnthropicConfig(AnthropicModelInfo, BaseConfig):
                 type="enabled",
                 budget_tokens=DEFAULT_REASONING_EFFORT_MINIMAL_THINKING_BUDGET,
             )
+        elif reasoning_effort == "xhigh":
+            return AnthropicThinkingParam(
+                type="enabled",
+                budget_tokens=DEFAULT_REASONING_EFFORT_XHIGH_THINKING_BUDGET,
+            )
+        elif reasoning_effort == "max":
+            return AnthropicThinkingParam(
+                type="enabled",
+                budget_tokens=DEFAULT_REASONING_EFFORT_MAX_THINKING_BUDGET,
+            )
         else:
             raise ValueError(f"Unmapped reasoning effort: {reasoning_effort}")
 
diff --git a/tests/test_litellm/llms/anthropic/chat/test_anthropic_chat_transformation.py b/tests/test_litellm/llms/anthropic/chat/test_anthropic_chat_transformation.py
index cbb9b21d620..b8f5ded1cd2 100644
--- a/tests/test_litellm/llms/anthropic/chat/test_anthropic_chat_transformation.py
+++ b/tests/test_litellm/llms/anthropic/chat/test_anthropic_chat_transformation.py
@@ -2159,6 +2159,8 @@ def test_reasoning_effort_maps_to_budget_thinking_for_non_opus_4_6():
         ("medium", 2048),  # DEFAULT_REASONING_EFFORT_MEDIUM_THINKING_BUDGET
         ("high", 4096),  # DEFAULT_REASONING_EFFORT_HIGH_THINKING_BUDGET
         ("minimal", 128),  # DEFAULT_REASONING_EFFORT_MINIMAL_THINKING_BUDGET
+        ("xhigh", 8192),  # DEFAULT_REASONING_EFFORT_XHIGH_THINKING_BUDGET
+        ("max", 16384),  # DEFAULT_REASONING_EFFORT_MAX_THINKING_BUDGET
     ]
 
     for effort, expected_budget in test_cases: