fix(bedrock): use tool fallback without forced tool_choice for claude-fable-5-1 structured output

Co-Authored-By: Devin AI <158243242+devin-ai-integration[bot]@users.noreply.github.com>

litellm/llms/bedrock/chat/converse_transformation.py

@@ -1073,6 +1073,7 @@ class AmazonConverseConfig(BaseConfig):
             if (
                 litellm.utils.supports_tool_choice(model=model, custom_llm_provider=self.custom_llm_provider)
                 and not is_thinking_enabled
+                and not AnthropicModelInfo.forced_tool_use_unsupported(model)
             ):
                 optional_params["tool_choice"] = ToolChoiceValuesBlock(
                     tool=SpecificToolChoiceBlock(name=RESPONSE_FORMAT_TOOL_NAME)

litellm/llms/bedrock/chat/invoke_transformations/anthropic_claude3_transformation.py

@@ -3,6 +3,7 @@ from typing import TYPE_CHECKING, Any, Final
 import httpx
 
 from litellm.anthropic_beta_headers_manager import filter_and_transform_beta_headers
+from litellm.constants import RESPONSE_FORMAT_TOOL_NAME
 from litellm.litellm_core_utils.prompt_templates.factory import (
     convert_to_anthropic_image_obj,
 )
@@ -11,6 +12,7 @@ from litellm.litellm_core_utils.prompt_templates.image_handling import (
     convert_url_to_base64,
 )
 from litellm.llms.anthropic.chat.transformation import AnthropicConfig
+from litellm.llms.anthropic.common_utils import AnthropicModelInfo
 from litellm.llms.bedrock.chat.invoke_transformations.base_invoke_transformation import (
     AmazonInvokeConfig,
 )
@@ -105,6 +107,16 @@ class AmazonAnthropicClaudeConfig(AmazonInvokeConfig, AnthropicConfig):
         # Restore original model name
         model = original_model
 
+        # The stub model hides the original model from the parent's forced-tool-use backstop
+        response_format_tool_choice: Final = optional_params.get("tool_choice")
+        if (
+            "response_format" in non_default_params
+            and isinstance(response_format_tool_choice, dict)
+            and response_format_tool_choice.get("name") == RESPONSE_FORMAT_TOOL_NAME
+            and AnthropicModelInfo.forced_tool_use_unsupported(original_model)
+        ):
+            optional_params.pop("tool_choice")
+
         return optional_params
 
     @staticmethod

litellm/model_prices_and_context_window_backup.json

@@ -1482,7 +1482,7 @@
         "supports_tool_choice": true,
         "supports_vision": true,
         "supports_xhigh_reasoning_effort": true,
-        "supports_native_structured_output": true,
+        "supports_native_structured_output": false,
         "supports_max_reasoning_effort": true,
         "supports_output_config": true,
         "bedrock_output_config_effort_ceiling": "xhigh",
@@ -1557,7 +1557,7 @@
         "supports_tool_choice": true,
         "supports_vision": true,
         "supports_xhigh_reasoning_effort": true,
-        "supports_native_structured_output": true,
+        "supports_native_structured_output": false,
         "supports_max_reasoning_effort": true,
         "supports_output_config": true,
         "bedrock_output_config_effort_ceiling": "xhigh",
@@ -1632,7 +1632,7 @@
         "supports_tool_choice": true,
         "supports_vision": true,
         "supports_xhigh_reasoning_effort": true,
-        "supports_native_structured_output": true,
+        "supports_native_structured_output": false,
         "supports_max_reasoning_effort": true,
         "supports_output_config": true,
         "bedrock_output_config_effort_ceiling": "xhigh",
@@ -1707,7 +1707,7 @@
         "supports_tool_choice": true,
         "supports_vision": true,
         "supports_xhigh_reasoning_effort": true,
-        "supports_native_structured_output": true,
+        "supports_native_structured_output": false,
         "supports_max_reasoning_effort": true,
         "supports_output_config": true,
         "bedrock_output_config_effort_ceiling": "xhigh",

model_prices_and_context_window.json

@@ -1482,7 +1482,7 @@
         "supports_tool_choice": true,
         "supports_vision": true,
         "supports_xhigh_reasoning_effort": true,
-        "supports_native_structured_output": true,
+        "supports_native_structured_output": false,
         "supports_max_reasoning_effort": true,
         "supports_output_config": true,
         "bedrock_output_config_effort_ceiling": "xhigh",
@@ -1557,7 +1557,7 @@
         "supports_tool_choice": true,
         "supports_vision": true,
         "supports_xhigh_reasoning_effort": true,
-        "supports_native_structured_output": true,
+        "supports_native_structured_output": false,
         "supports_max_reasoning_effort": true,
         "supports_output_config": true,
         "bedrock_output_config_effort_ceiling": "xhigh",
@@ -1632,7 +1632,7 @@
         "supports_tool_choice": true,
         "supports_vision": true,
         "supports_xhigh_reasoning_effort": true,
-        "supports_native_structured_output": true,
+        "supports_native_structured_output": false,
         "supports_max_reasoning_effort": true,
         "supports_output_config": true,
         "bedrock_output_config_effort_ceiling": "xhigh",
@@ -1707,7 +1707,7 @@
         "supports_tool_choice": true,
         "supports_vision": true,
         "supports_xhigh_reasoning_effort": true,
-        "supports_native_structured_output": true,
+        "supports_native_structured_output": false,
         "supports_max_reasoning_effort": true,
         "supports_output_config": true,
         "bedrock_output_config_effort_ceiling": "xhigh",

tests/test_litellm/llms/bedrock/chat/invoke_transformations/test_bedrock_chat_invoke_transformations_anthropic_claude3_transformation.py

@@ -649,9 +649,9 @@ def test_bedrock_chat_invoke_drop_params_still_inlines_for_non_native(local_mode
     "model",
     ["us.anthropic.claude-fable-5-1", "anthropic.claude-fable-5-1"],
 )
-def test_bedrock_chat_invoke_fable_5_1_response_format_uses_native_path(local_model_cost_map, model):
-    """Regression: Fable 5.1 rejects forced tool use, so invoke must skip the
-    tool-based structured-output stub and emit ``output_format`` instead of a
+def test_bedrock_chat_invoke_fable_5_1_response_format_avoids_forced_tool_choice(local_model_cost_map, model):
+    """Regression: Bedrock rejects both native ``output_config.format`` and forced
+    tool_choice for Fable 5.1, so invoke must use the tool-based path without a
     forced ``tool_choice``."""
     result = AmazonAnthropicClaudeConfig().map_openai_params(
         non_default_params={
@@ -668,5 +668,6 @@ def test_bedrock_chat_invoke_fable_5_1_response_format_uses_native_path(local_mo
         drop_params=False,
     )
 
-    assert "output_format" in result
+    assert "output_format" not in result
+    assert "tools" in result
     assert "tool_choice" not in result

tests/test_litellm/llms/bedrock/chat/test_converse_transformation.py

@@ -6497,6 +6497,39 @@ def test_unforced_tool_choice_unaffected_on_fable_5_1_converse(local_model_cost_
     assert result == ({"auto": {}} if tool_choice == "auto" else None)
 
 
+@pytest.mark.parametrize(
+    "model",
+    ["anthropic.claude-fable-5-1", "us.anthropic.claude-fable-5-1"],
+)
+def test_response_format_avoids_native_and_forced_tool_choice_on_fable_5_1_converse(
+    local_model_cost_map, model
+):
+    """Regression: Bedrock rejects both ``outputConfig`` structured output and forced
+    tool_choice for Fable 5.1, so response_format must map to a tool without a forced
+    tool_choice."""
+    config = AmazonConverseConfig()
+
+    result = config.map_openai_params(
+        non_default_params={
+            "response_format": {
+                "type": "json_schema",
+                "json_schema": {
+                    "name": "test_schema",
+                    "schema": {"type": "object", "properties": {"result": {"type": "string"}}},
+                },
+            }
+        },
+        optional_params={},
+        model=model,
+        drop_params=False,
+    )
+
+    assert "outputConfig" not in result
+    assert "tools" in result
+    assert "tool_choice" not in result
+    assert result.get("json_mode") is True
+
+
 def test_forced_tool_choice_forwarded_on_converse_models_that_support_it(
     local_model_cost_map, monkeypatch
 ):