From 4fdbbdfe6de73637e631ff41ad74da8f6e104eab Mon Sep 17 00:00:00 2001
From: Cesar Garcia <128240629+Chesars@users.noreply.github.com>
Date: Mon, 15 Dec 2025 22:57:40 -0300
Subject: [PATCH] fix(anthropic): correct claude-3-7-sonnet max_tokens to 64K
 default (#17979)

Claude 3.7 Sonnet's default max_output_tokens is 64000, not 128000.
The 128K output limit requires the beta header 'output-128k-2025-02-19'.

This fixes the integration test failure where requests with max_tokens=128000
were being rejected by the Anthropic API.

Fixes test_multiturn_tool_calls in test_anthropic_responses_api.py
---
 litellm/model_prices_and_context_window_backup.json   |  8 ++++----
 model_prices_and_context_window.json                  |  8 ++++----
 .../chat/test_anthropic_chat_transformation.py        | 11 ++++++-----
 3 files changed, 14 insertions(+), 13 deletions(-)

diff --git a/litellm/model_prices_and_context_window_backup.json b/litellm/model_prices_and_context_window_backup.json
index 4b016bc6ca..26aae425fa 100644
--- a/litellm/model_prices_and_context_window_backup.json
+++ b/litellm/model_prices_and_context_window_backup.json
@@ -6723,8 +6723,8 @@
         "input_cost_per_token": 3e-06,
         "litellm_provider": "anthropic",
         "max_input_tokens": 200000,
-        "max_output_tokens": 128000,
-        "max_tokens": 128000,
+        "max_output_tokens": 64000,
+        "max_tokens": 64000,
         "mode": "chat",
         "output_cost_per_token": 1.5e-05,
         "search_context_cost_per_query": {
@@ -6752,8 +6752,8 @@
         "input_cost_per_token": 3e-06,
         "litellm_provider": "anthropic",
         "max_input_tokens": 200000,
-        "max_output_tokens": 128000,
-        "max_tokens": 128000,
+        "max_output_tokens": 64000,
+        "max_tokens": 64000,
         "mode": "chat",
         "output_cost_per_token": 1.5e-05,
         "search_context_cost_per_query": {
diff --git a/model_prices_and_context_window.json b/model_prices_and_context_window.json
index 4b016bc6ca..26aae425fa 100644
--- a/model_prices_and_context_window.json
+++ b/model_prices_and_context_window.json
@@ -6723,8 +6723,8 @@
         "input_cost_per_token": 3e-06,
         "litellm_provider": "anthropic",
         "max_input_tokens": 200000,
-        "max_output_tokens": 128000,
-        "max_tokens": 128000,
+        "max_output_tokens": 64000,
+        "max_tokens": 64000,
         "mode": "chat",
         "output_cost_per_token": 1.5e-05,
         "search_context_cost_per_query": {
@@ -6752,8 +6752,8 @@
         "input_cost_per_token": 3e-06,
         "litellm_provider": "anthropic",
         "max_input_tokens": 200000,
-        "max_output_tokens": 128000,
-        "max_tokens": 128000,
+        "max_output_tokens": 64000,
+        "max_tokens": 64000,
         "mode": "chat",
         "output_cost_per_token": 1.5e-05,
         "search_context_cost_per_query": {
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 ec612109d9..9b6d1c6e17 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
@@ -1651,15 +1651,16 @@ def test_get_max_tokens_for_model_claude_35():
 def test_get_max_tokens_for_model_claude_37():
     """
     Test that get_max_tokens_for_model returns correct value for Claude 3.7 models.
-    Claude 3.7 Sonnet has max_output_tokens of 128000 (128K with extended thinking).
+    Claude 3.7 Sonnet has max_output_tokens of 64000 by default.
+    128K output requires the beta header 'output-128k-2025-02-19'.
 
     Fixes: https://github.com/BerriAI/litellm/issues/8835
     """
     config = AnthropicConfig()
 
-    # Claude 3.7 Sonnet should return 128000 (128K)
+    # Claude 3.7 Sonnet should return 64000 (64K default, 128K requires beta header)
     max_tokens = config.get_max_tokens_for_model("claude-3-7-sonnet-20250219")
-    assert max_tokens == 128000
+    assert max_tokens == 64000
 
 
 def test_get_max_tokens_for_model_unknown():
@@ -1698,9 +1699,9 @@ def test_get_config_with_model_uses_dynamic_max_tokens():
     config_claude35 = AnthropicConfig.get_config(model="claude-3-5-sonnet-20241022")
     assert config_claude35["max_tokens"] == 8192
 
-    # Claude 3.7 model should get 128000 (128K with extended thinking)
+    # Claude 3.7 model should get 64000 (64K default, 128K requires beta header)
     config_claude37 = AnthropicConfig.get_config(model="claude-3-7-sonnet-20250219")
-    assert config_claude37["max_tokens"] == 128000
+    assert config_claude37["max_tokens"] == 64000
 
 
 def test_get_config_without_model_uses_fallback():