diff --git a/litellm/utils.py b/litellm/utils.py
index 3439a21b560..f3b9fcfd1ed 100644
--- a/litellm/utils.py
+++ b/litellm/utils.py
@@ -5624,6 +5624,12 @@ def _get_model_info_from_generalization(
     return None
 
 
+def _strip_mantle_region_prefix(model: str) -> str:
+    from litellm.llms.bedrock_mantle.common_utils import split_mantle_region_prefix
+
+    return split_mantle_region_prefix(model)[1]
+
+
 def _get_potential_model_names(model: str, custom_llm_provider: str | None) -> PotentialModelNamesAndCustomLLMProvider:
     if custom_llm_provider is None:
         # Get custom_llm_provider
@@ -5656,17 +5662,22 @@ def _get_potential_model_names(model: str, custom_llm_provider: str | None) -> P
 
         split_model = strip_bedrock_routing_prefix(split_model)
 
+    region_free_split_model: Final = (
+        _strip_mantle_region_prefix(split_model) if custom_llm_provider == "bedrock_mantle" else split_model
+    )
     provider_model_info: Final = (
-        ProviderConfigManager.get_provider_model_info(model=split_model, provider=LlmProviders(custom_llm_provider))
+        ProviderConfigManager.get_provider_model_info(
+            model=region_free_split_model, provider=LlmProviders(custom_llm_provider)
+        )
         if custom_llm_provider in LlmProvidersSet
         else None
     )
     provider_cost_key: Final = (
-        provider_model_info.get_model_cost_key(split_model) if provider_model_info is not None else None
+        provider_model_info.get_model_cost_key(region_free_split_model) if provider_model_info is not None else None
     )
 
     return PotentialModelNamesAndCustomLLMProvider(
-        split_model=split_model,
+        split_model=region_free_split_model,
         combined_model_name=combined_model_name,
         stripped_model_name=stripped_model_name,
         combined_stripped_model_name=combined_stripped_model_name,
diff --git a/tests/test_litellm/test_cost_calculator.py b/tests/test_litellm/test_cost_calculator.py
index aef17f3d5d0..fe52b9993f2 100644
--- a/tests/test_litellm/test_cost_calculator.py
+++ b/tests/test_litellm/test_cost_calculator.py
@@ -3522,6 +3522,31 @@ def test_cost_per_token_region_name_applies_to_provider_prefixed_model(_local_mo
     )
 
 
+def test_completion_cost_mantle_native_messages_prices_claude_from_the_bedrock_row(_local_model_cost_map):
+    """Mantle's native Messages API answers with Anthropic's canonical model name and the proxy
+    resolves a Mantle region for every call, so the first cost candidate is
+    bedrock_mantle/<region>/claude-sonnet-5. That name has no row of its own and must fall through to
+    the deployment's bare Bedrock row instead of stopping on an unpriced capability rule at $0."""
+
+    response = litellm.ModelResponse(
+        id="msg_x",
+        choices=[{"index": 0, "message": {"role": "assistant", "content": "hi"}, "finish_reason": "stop"}],
+        model="claude-sonnet-5",
+        usage={"prompt_tokens": 100, "completion_tokens": 10, "total_tokens": 110},
+    )
+    row = litellm.model_cost["anthropic.claude-sonnet-5"]
+    expected = 100 * row["input_cost_per_token"] + 10 * row["output_cost_per_token"]
+    assert expected > 0
+
+    for region_name in ("us-east-1", None):
+        assert litellm.completion_cost(
+            completion_response=response,
+            model="bedrock_mantle/anthropic.claude-sonnet-5",
+            custom_llm_provider="bedrock_mantle",
+            region_name=region_name,
+        ) == pytest.approx(expected)
+
+
 def test_select_model_name_keeps_base_model_free_of_region(_local_model_cost_map):
     """An explicit base_model keeps pricing on that model's own key even when the request carries a
     region with different regional rates, so the private provider model never widens region pricing."""