fix(bedrock): price bedrock/mantle/<model> deployments from the base model row (#42402)

Co-authored-by: mateo-berri <277851410+mateo-berri@users.noreply.github.com>

litellm/llms/bedrock/common_utils.py

@@ -787,7 +787,7 @@ def is_bedrock_application_inference_profile_arn(model: str) -> bool:
 
 def strip_bedrock_routing_prefix(model: str) -> str:
     """Strip LiteLLM routing prefixes from model name."""
-    for prefix in ["bedrock/", "converse/", "invoke/", "openai/", "nova-2/", "nova/"]:
+    for prefix in ["bedrock/", "converse/", "invoke/", "openai/", "mantle/", "nova-2/", "nova/"]:
         if model.startswith(prefix):
             model = model.split("/", 1)[1]
     return model
@@ -850,6 +850,7 @@ def get_bedrock_base_model(model: str) -> str:
     Handle model names like:
     - "us.meta.llama3-2-11b-instruct-v1:0" -> "meta.llama3-2-11b-instruct-v1"
     - "bedrock/converse/model" -> "model"
+    - "bedrock/mantle/anthropic.claude-sonnet-5" -> "anthropic.claude-sonnet-5"
     - "anthropic.claude-3-5-sonnet-20241022-v2:0:51k" -> "anthropic.claude-3-5-sonnet-20241022-v2:0"
     - "bedrock/nova-2/arn:aws:..." -> "amazon.nova-2-custom"
     - "bedrock/nova/arn:aws:..." -> "amazon.nova-custom"

tests/test_litellm/llms/bedrock/test_bedrock_common_utils.py

@@ -279,6 +279,16 @@ def test_context_window_suffix_stripped_for_cost_lookup():
     )
 
 
+def test_legacy_mantle_route_prefix_stripped_for_cost_lookup():
+    """The mantle/ route token is a routing prefix like openai/, so a bedrock/mantle/<model>
+    deployment must resolve the bare Bedrock model for cost lookup while still routing to Mantle."""
+    from litellm.llms.bedrock.common_utils import get_bedrock_base_model, strip_bedrock_routing_prefix
+
+    assert strip_bedrock_routing_prefix("mantle/anthropic.claude-sonnet-5") == "anthropic.claude-sonnet-5"
+    assert get_bedrock_base_model("bedrock/mantle/anthropic.claude-sonnet-5") == "anthropic.claude-sonnet-5"
+    assert BedrockModelInfo.get_bedrock_route("bedrock/mantle/anthropic.claude-sonnet-5") == "mantle"
+
+
 def test_output_config_effort_normalization_uses_model_info_ceiling(monkeypatch):
     import litellm.llms.bedrock.common_utils as mod
 

tests/test_litellm/test_cost_calculator.py

@@ -3574,6 +3574,45 @@ def test_completion_cost_mantle_native_messages_prices_haiku_from_the_mantle_row
         ) == pytest.approx(expected), model
 
 
+def test_completion_cost_legacy_mantle_route_prices_after_router_registration(local_model_cost_map):
+    """The proxy registers every deployment under its provider-prefixed key at boot. A
+    bedrock/mantle/<model> deployment must resolve to the bare Bedrock row there, otherwise the boot
+    entry is a cost-less capability rule that shadows the priced row and every call on the deployment,
+    /v1/chat/completions and /v1/messages alike, bills $0."""
+    from litellm import Router
+
+    Router(
+        model_list=[
+            {
+                "model_name": "claude-sonnet-5",
+                "litellm_params": {
+                    "model": "bedrock/mantle/anthropic.claude-sonnet-5",
+                    "aws_region_name": "us-east-1",
+                },
+            }
+        ]
+    )
+    assert "bedrock/mantle/anthropic.claude-sonnet-5" not in litellm.model_cost
+
+    response = litellm.ModelResponse(
+        id="msg_x",
+        choices=[{"index": 0, "message": {"role": "assistant", "content": "hi"}, "finish_reason": "stop"}],
+        model="claude-sonnet-5",
+        usage={"prompt_tokens": 16, "completion_tokens": 4, "total_tokens": 20},
+    )
+    row = litellm.model_cost["anthropic.claude-sonnet-5"]
+    expected = 16 * row["input_cost_per_token"] + 4 * row["output_cost_per_token"]
+    assert expected > 0
+
+    for call_type in ("completion", "anthropic_messages"):
+        assert litellm.completion_cost(
+            completion_response=response,
+            model="mantle/anthropic.claude-sonnet-5",
+            custom_llm_provider="bedrock",
+            call_type=call_type,
+        ) == pytest.approx(expected), call_type
+
+
 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."""