fix(bedrock): pre-clamp reasoning_effort for chat invoke; correct test caps

cursoragent · yassin-berriai · cursoragent · commit d8c24e8260ad · 2026-05-28T00:59:52.000Z
- Add _clamp_adaptive_reasoning_effort_for_bedrock to AmazonAnthropicClaudeConfig
  so raw reasoning_effort=xhigh degrades to the model's bedrock effort ceiling
  before AnthropicConfig.map_openai_params converts it to output_config.
  Mirrors converse path (_handle_reasoning_effort_parameter) and messages path
  (_clamp_adaptive_reasoning_effort_for_bedrock) so the three Bedrock paths
  are consistent.

- grid_spec: restore caps=_CAPS_4_6 for Bedrock converse/invoke Opus 4.6 entries
  so the test reflects the model's actual JSON capabilities. Teach expected()
  to bypass the xhigh/max cap check when bedrock_effort_ceiling will clamp
  the wire effort, so the test still passes for Bedrock's graceful degradation
  contract without lying about native model caps.

Co-authored-by: Yassin Kortam &lt;yassin@berri.ai&gt;
diff --git a/litellm/llms/bedrock/chat/invoke_transformations/anthropic_claude3_transformation.py b/litellm/llms/bedrock/chat/invoke_transformations/anthropic_claude3_transformation.py
@@ -78,6 +78,17 @@ def map_openai_params(
             # Use a model name that forces tool-based approach
             model = "claude-3-sonnet-20240229"
 
+        # Clamp ``reasoning_effort`` to the Bedrock effort ceiling before the
+        # parent mapping converts it to ``output_config.effort`` and the
+        # downstream effort gate runs. Mirrors the converse path's
+        # ``_handle_reasoning_effort_parameter`` and the messages path's
+        # ``_clamp_adaptive_reasoning_effort_for_bedrock`` so adaptive Claude
+        # requests degrade ``xhigh`` -> ``max`` rather than 400-ing on
+        # models like Opus 4.6 that don't natively advertise xhigh.
+        self._clamp_adaptive_reasoning_effort_for_bedrock(
+            model=original_model, params=non_default_params
+        )
+
         optional_params = AnthropicConfig.map_openai_params(
             self,
             non_default_params,
@@ -91,6 +102,27 @@ def map_openai_params(
 
         return optional_params
 
+    @staticmethod
+    def _clamp_adaptive_reasoning_effort_for_bedrock(model: str, params: dict) -> None:
+        """Lower ``reasoning_effort`` to the Bedrock effort ceiling before mapping.
+
+        Bedrock's adaptive Claude models accept the OpenAI-style
+        ``reasoning_effort`` tier, but the request validator can reject tiers
+        the model does not natively advertise (e.g. ``xhigh`` on Opus 4.6).
+        Clamp the raw tier to the model's
+        ``bedrock_output_config_effort_ceiling`` so Claude Code "goal mode"
+        keeps working. Non-adaptive models and models without a ceiling are
+        left untouched.
+        """
+        if not AnthropicConfig._is_adaptive_thinking_model(model):
+            return
+        effort = params.get("reasoning_effort")
+        if not isinstance(effort, str):
+            return
+        clamped = {"effort": effort}
+        normalize_bedrock_opus_output_config_effort(model=model, output_config=clamped)
+        params["reasoning_effort"] = clamped["effort"]
+
     def transform_request(
         self,
         model: str,
diff --git a/tests/llm_translation/reasoning_effort_grid/grid_spec.py b/tests/llm_translation/reasoning_effort_grid/grid_spec.py
@@ -71,6 +71,20 @@ def params(self) -> Dict[str, str]:
 _BAD_REQUEST_EFFORTS: FrozenSet[str] = frozenset({"disabled", "invalid", ""})
 
 
+def _bedrock_clamps_effort(model: "ModelEntry", effort: str) -> bool:
+    """Whether Bedrock will clamp ``effort`` down to ``bedrock_effort_ceiling``.
+
+    Bedrock chat/messages paths clamp unsupported high tiers (e.g. ``xhigh``
+    on Opus 4.6) to the model's ceiling rather than rejecting them, so the
+    missing native capability is OK — the wire effort just degrades.
+    """
+    if model.bedrock_effort_ceiling is None:
+        return False
+    if effort not in _EFFORT_RANK or model.bedrock_effort_ceiling not in _EFFORT_RANK:
+        return False
+    return _EFFORT_RANK[effort] > _EFFORT_RANK[model.bedrock_effort_ceiling]
+
+
 def expected(model: ModelEntry, effort: str) -> CellExpectation:
     if effort in ("__omit__", "none"):
         if model.mode == "budget":
@@ -82,7 +96,7 @@ def expected(model: ModelEntry, effort: str) -> CellExpectation:
 
     if effort in ("xhigh", "max"):
         cap = f"supports_{effort}_reasoning_effort"
-        if cap not in model.caps:
+        if cap not in model.caps and not _bedrock_clamps_effort(model, effort):
             return CellExpectation(status=400, thinking_type=OMIT)
 
     if model.mode == "adaptive":
@@ -233,7 +247,7 @@ def expected(model: ModelEntry, effort: str) -> CellExpectation:
         mode="adaptive",
         extra_params=(("aws_region_name", "us-east-1"),),
         required_env=_BEDROCK_REQ,
-        caps=_CAPS_OPUS_4_7,
+        caps=_CAPS_4_6,
         bedrock_effort_ceiling="max",
     ),
     ModelEntry(
@@ -262,7 +276,7 @@ def expected(model: ModelEntry, effort: str) -> CellExpectation:
         mode="adaptive",
         extra_params=(("aws_region_name", "us-east-1"),),
         required_env=_BEDROCK_REQ,
-        caps=_CAPS_OPUS_4_7,
+        caps=_CAPS_4_6,
         bedrock_effort_ceiling="max",
     ),
     ModelEntry(
