Merge remote-tracking branch 'origin/litellm_internal_staging' into litellm_/modest-hodgkin-5775d7

ci_cd/generate_model_prices_schema.py

@@ -41,6 +41,11 @@ OBJECT_KEYS: dict[str, JsonSchema] = {
         },
         "additionalProperties": False,
     },
+    "guardrail_cost_per_unit": {
+        "type": "object",
+        "description": "USD cost per billable guardrail unit, keyed by the provider's usage counter name (e.g. Bedrock's contentPolicyUnits).",
+        "additionalProperties": NONNEG_NUMBER,
+    },
     "metadata": {
         "type": "object",
         "description": "Free-form notes about the entry (e.g. pricing derivation).",

litellm/__init__.py

@@ -792,6 +792,8 @@ def _populate_provider_model_sets(model_cost_map: Dict) -> None:
             nlp_cloud_models.add(key)
         elif value.get("litellm_provider") == "aleph_alpha":
             aleph_alpha_models.add(key)
+        elif value.get("litellm_provider") == "bedrock" and value.get("mode") == "guardrail":
+            pass
         elif value.get("litellm_provider") == "bedrock" and not is_bedrock_pricing_only_model(key):
             bedrock_models.add(key)
         elif value.get("litellm_provider") == "bedrock_converse":

litellm/litellm_core_utils/litellm_logging.py

@@ -64,6 +64,10 @@ from litellm.integrations.mlflow import MlflowLogger
 from litellm.integrations.sqs import SQSLogger
 from litellm.litellm_core_utils.core_helpers import reconstruct_model_name
 from litellm.litellm_core_utils.get_litellm_params import get_litellm_params
+from litellm.litellm_core_utils.llm_cost_calc.guardrail_cost import (
+    cost_breakdown_with_guardrail,
+    guardrail_information_cost,
+)
 from litellm.litellm_core_utils.llm_cost_calc.tool_call_cost_tracking import (
     StandardBuiltInToolCostTracking,
 )
@@ -5650,12 +5654,14 @@ def get_standard_logging_object_payload(
             base_model = metadata.get("deployment")
         custom_pricing: Final = use_custom_pricing_for_model(litellm_params=litellm_params)
         raw_response_cost: Final = kwargs.get("response_cost")
-        response_cost: Final[float] = raw_response_cost or 0.0
+        llm_response_cost: Final[float] = raw_response_cost or 0.0
+        guardrail_cost: Final = guardrail_information_cost(metadata.get("standard_logging_guardrail_information"))
+        response_cost: Final[float] = llm_response_cost + guardrail_cost
 
         # clean up litellm hidden params
         clean_hidden_params: Final = StandardLoggingPayloadSetup.get_hidden_params(hidden_params)
         if clean_hidden_params["response_cost"] is None and raw_response_cost is not None:
-            clean_hidden_params["response_cost"] = response_cost
+            clean_hidden_params["response_cost"] = llm_response_cost
 
         model_cost_information: Final = StandardLoggingPayloadSetup.get_model_cost_information(
             base_model=base_model,
@@ -5735,7 +5741,7 @@ def get_standard_logging_object_payload(
             metadata=clean_metadata,
             cache_key=clean_hidden_params["cache_key"],
             response_cost=response_cost,
-            cost_breakdown=logging_obj.cost_breakdown,
+            cost_breakdown=cost_breakdown_with_guardrail(logging_obj.cost_breakdown, guardrail_cost),
             total_tokens=usage_dict.get("total_tokens", 0),
             prompt_tokens=usage_dict.get("prompt_tokens", 0),
             completion_tokens=usage_dict.get("completion_tokens", 0),

litellm/litellm_core_utils/llm_cost_calc/guardrail_cost.py

@@ -0,0 +1,78 @@
+import math
+from collections.abc import Mapping
+from typing import Final
+
+from pydantic import BaseModel, ConfigDict, TypeAdapter, ValidationError
+
+import litellm
+from litellm._logging import verbose_logger
+from litellm.types.utils import CostBreakdown
+
+BEDROCK_GUARDRAIL_PRICING_KEY: Final = "bedrock/guardrails"
+
+
+class GuardrailPricing(BaseModel):
+    model_config = ConfigDict(extra="ignore", frozen=True)
+
+    guardrail_cost_per_unit: Mapping[str, float]
+
+
+class GuardrailCostEntry(BaseModel):
+    model_config = ConfigDict(extra="ignore", frozen=True)
+
+    guardrail_cost: float | None = None
+
+
+GuardrailInformationShape = tuple[GuardrailCostEntry, ...] | GuardrailCostEntry | None
+
+_GUARDRAIL_INFORMATION_ADAPTER: Final[TypeAdapter[GuardrailInformationShape]] = TypeAdapter(GuardrailInformationShape)
+
+
+def _bedrock_guardrail_pricing(aws_region_name: str | None) -> GuardrailPricing | None:
+    regional_key: Final = f"bedrock/{aws_region_name}/guardrails" if aws_region_name else None
+    for key in (regional_key, BEDROCK_GUARDRAIL_PRICING_KEY):
+        if key is None or key not in litellm.model_cost:
+            continue
+        try:
+            return GuardrailPricing.model_validate(litellm.model_cost[key])
+        except ValidationError as e:
+            verbose_logger.warning("Ignoring malformed guardrail pricing entry %s: %s", key, e)
+    return None
+
+
+def bedrock_guardrail_cost(usage_units: Mapping[str, int], aws_region_name: str | None) -> float:
+    pricing: Final = _bedrock_guardrail_pricing(aws_region_name)
+    if pricing is None:
+        return 0.0
+    return sum(units * pricing.guardrail_cost_per_unit.get(counter, 0.0) for counter, units in usage_units.items())
+
+
+def _billable_entry_cost(entry: GuardrailCostEntry) -> float:
+    cost: Final = entry.guardrail_cost
+    if cost is None or not math.isfinite(cost) or cost <= 0.0:
+        return 0.0
+    return cost
+
+
+def guardrail_information_cost(guardrail_information: object) -> float:
+    try:
+        parsed: Final = _GUARDRAIL_INFORMATION_ADAPTER.validate_python(guardrail_information)
+    except ValidationError:
+        return 0.0
+    if parsed is None:
+        return 0.0
+    if isinstance(parsed, GuardrailCostEntry):
+        return _billable_entry_cost(parsed)
+    return sum(_billable_entry_cost(entry) for entry in parsed)
+
+
+def cost_breakdown_with_guardrail(cost_breakdown: CostBreakdown | None, guardrail_cost: float) -> CostBreakdown | None:
+    if guardrail_cost <= 0.0:
+        return cost_breakdown
+    existing: Final[CostBreakdown] = cost_breakdown if cost_breakdown is not None else CostBreakdown()
+    merged: Final[CostBreakdown] = {
+        **existing,
+        "guardrail_cost": guardrail_cost,
+        "total_cost": existing.get("total_cost", 0.0) + guardrail_cost,
+    }
+    return merged

litellm/litellm_core_utils/llm_cost_calc/utils.py

@@ -42,14 +42,18 @@ _VALID_DATA_RESIDENCIES: Final = frozenset(r.value for r in DataResidency)
 
 # Pre-resolved service-tier cost-key suffixes (e.g. "_priority"). Used per
 # request in the cost-calc path, so the f-strings are built once here instead
-# of being rebuilt for every model_info key on every call.
-_SERVICE_TIER_SUFFIXES: Final[tuple[str, ...]] = tuple(f"_{st.value}" for st in ServiceTier)
+# of being rebuilt for every model_info key on every call. Longest-first so a
+# substring match resolves "_ultrafast" before "_fast".
+_SERVICE_TIER_SUFFIXES: Final[tuple[str, ...]] = tuple(
+    sorted((f"_{st.value}" for st in ServiceTier), key=len, reverse=True)
+)
 
 _SERVICE_TIER_TO_COST_KEY_SUFFIX: Final[Mapping[str, str]] = MappingProxyType(
     {
         ServiceTier.FLEX.value: ServiceTier.FLEX.value,
         ServiceTier.PRIORITY.value: ServiceTier.PRIORITY.value,
         ServiceTier.FAST.value: ServiceTier.PRIORITY.value,
+        ServiceTier.ULTRAFAST.value: ServiceTier.ULTRAFAST.value,
     }
 )
 
@@ -191,7 +195,7 @@ def _get_service_tier_cost_key(base_key: str, service_tier: str | None) -> str:
 
     Args:
         base_key: The base cost key (e.g., "input_cost_per_token")
-        service_tier: The service tier ("flex", "priority", "fast", or None for standard)
+        service_tier: The service tier ("flex", "priority", "fast", "ultrafast", or None for standard)
 
     Returns:
         str: The cost key to use (e.g., "input_cost_per_token_flex" or "input_cost_per_token")

litellm/model_prices_and_context_window_backup.json

@@ -10052,6 +10052,21 @@
         "output_cost_per_second": 0.0066027,
         "supports_tool_choice": true
     },
+    "bedrock/guardrails": {
+        "guardrail_cost_per_unit": {
+            "automatedReasoningPolicyUnits": 0.00017,
+            "contentPolicyImageUnits": 0.00075,
+            "contentPolicyUnits": 0.00015,
+            "contextualGroundingPolicyUnits": 0.0001,
+            "sensitiveInformationPolicyFreeUnits": 0.0,
+            "sensitiveInformationPolicyUnits": 0.0001,
+            "topicPolicyUnits": 0.00015,
+            "wordPolicyUnits": 0.0
+        },
+        "litellm_provider": "bedrock",
+        "mode": "guardrail",
+        "source": "https://aws.amazon.com/bedrock/pricing/"
+    },
     "bedrock/ap-northeast-1/1-month-commitment/anthropic.claude-instant-v1": {
         "input_cost_per_second": 0.01475,
         "litellm_provider": "bedrock",

litellm/proxy/common_request_processing.py

@@ -32,11 +32,13 @@ from litellm.constants import (
     UNSAFE_PROXY_RESPONSE_HEADERS,
 )
 from litellm.integrations.custom_guardrail import CustomGuardrail
+from litellm.litellm_core_utils.core_helpers import get_or_create_metadata_bucket
 from litellm.litellm_core_utils.dd_tracing import NullTracer, tracer
 from litellm.litellm_core_utils.get_supported_openai_params import (
     get_supported_openai_params,
 )
 from litellm.litellm_core_utils.litellm_logging import Logging as LiteLLMLoggingObj
+from litellm.litellm_core_utils.llm_cost_calc.guardrail_cost import guardrail_information_cost
 from litellm.litellm_core_utils.llm_response_utils.get_headers import (
     get_response_headers,
 )
@@ -2462,11 +2464,21 @@ class ProxyBaseLLMRequestProcessing:
         additional_headers = hidden_params.get("additional_headers", {}) or {}
 
         recover_response_cost: Final = not response_cost and hidden_params.get("response_cost") is None
-        response_cost_for_headers: Final = (
+        llm_cost_for_headers: Final = (
             self._response_cost_from_logging_obj(response=response, logging_obj=logging_obj) or ""
             if recover_response_cost
             else response_cost
         )
+        _, request_metadata_bucket = get_or_create_metadata_bucket(self.data)
+        guardrail_cost_for_headers: Final = guardrail_information_cost(
+            request_metadata_bucket.get("standard_logging_guardrail_information")
+        )
+        response_cost_for_headers: Final = (
+            (llm_cost_for_headers if isinstance(llm_cost_for_headers, (int, float)) else 0.0)
+            + guardrail_cost_for_headers
+            if guardrail_cost_for_headers > 0
+            else llm_cost_for_headers
+        )
 
         fastapi_response.headers.update(
             ProxyBaseLLMRequestProcessing.get_custom_headers(

litellm/proxy/guardrails/guardrail_hooks/bedrock_guardrails.py

@@ -31,6 +31,7 @@ from litellm.constants import BEDROCK_APPLY_GUARDRAIL_CHUNK_BUDGET_CHARS
 from litellm.exceptions import ModifyResponseException
 from litellm.integrations.custom_guardrail import CustomGuardrail
 from litellm.litellm_core_utils.core_helpers import redact_nested_match_and_regex_keys
+from litellm.litellm_core_utils.llm_cost_calc.guardrail_cost import bedrock_guardrail_cost
 from litellm.llms.anthropic.chat.guardrail_translation.handler import AnthropicMessagesHandler
 from litellm.llms.base_llm.guardrail_translation.utils import (
     effective_scan_only_tool_results_for_guardrail,
@@ -872,6 +873,7 @@ class BedrockGuardrail(CustomGuardrail, BaseAWSLLM):
         credentials, aws_region_name = self._load_credentials()
         allow_chunking: Final = not self._content_uses_contextual_grounding(content)
 
+        completed_chunk_usages: Final[list[BedrockGuardrailUsage]] = []  # mutable-ok: billed-chunk usage accumulator
         try:
             responses: Final = await self._apply_guardrail_content_with_chunking(
                 content=content,
@@ -883,6 +885,7 @@ class BedrockGuardrail(CustomGuardrail, BaseAWSLLM):
                 event_type=event_type,
                 start_time=start_time,
                 allow_chunking=allow_chunking,
+                completed_chunk_usages=completed_chunk_usages,
             )
         except HTTPException as exc:
             if not isinstance(exc.detail, dict):
@@ -891,6 +894,8 @@ class BedrockGuardrail(CustomGuardrail, BaseAWSLLM):
                     request_data=request_data,
                     event_type=event_type,
                     start_time=start_time,
+                    aws_region_name=aws_region_name,
+                    completed_chunk_usages=completed_chunk_usages,
                 )
             raise
         merged_response: Final = self._merge_bedrock_guardrail_responses(responses)
@@ -899,6 +904,7 @@ class BedrockGuardrail(CustomGuardrail, BaseAWSLLM):
             request_data=request_data,
             event_type=event_type,
             start_time=start_time,
+            aws_region_name=aws_region_name,
         )
         return merged_response
 
@@ -913,6 +919,7 @@ class BedrockGuardrail(CustomGuardrail, BaseAWSLLM):
         event_type: GuardrailEventHooks,
         start_time: "datetime",
         allow_chunking: bool,
+        completed_chunk_usages: list[BedrockGuardrailUsage],  # mutable-ok: billed-chunk usage accumulator
     ) -> tuple[BedrockContentChunkResult, ...]:
         """Post `content` to ApplyGuardrail, chunking only if AWS rejects it as too large.
 
@@ -959,6 +966,7 @@ class BedrockGuardrail(CustomGuardrail, BaseAWSLLM):
                 request_data=request_data,
                 event_type=event_type,
                 start_time=start_time,
+                completed_chunk_usages=completed_chunk_usages,
             )
             return (
                 BedrockContentChunkResult(
@@ -989,6 +997,7 @@ class BedrockGuardrail(CustomGuardrail, BaseAWSLLM):
                             event_type=event_type,
                             start_time=start_time,
                             allow_chunking=allow_chunking,
+                            completed_chunk_usages=completed_chunk_usages,
                         )
                         for batch in batches
                     ]
@@ -1015,6 +1024,7 @@ class BedrockGuardrail(CustomGuardrail, BaseAWSLLM):
                     event_type=event_type,
                     start_time=start_time,
                     allow_chunking=allow_chunking,
+                    completed_chunk_usages=completed_chunk_usages,
                 )
                 second_results: Final = await self._apply_guardrail_content_with_chunking(
                     content=second_half,
@@ -1026,6 +1036,7 @@ class BedrockGuardrail(CustomGuardrail, BaseAWSLLM):
                     event_type=event_type,
                     start_time=start_time,
                     allow_chunking=allow_chunking,
+                    completed_chunk_usages=completed_chunk_usages,
                 )
                 combined_results: Final = tuple(first_results) + tuple(second_results)
                 if is_single_item_text_split:
@@ -1045,6 +1056,7 @@ class BedrockGuardrail(CustomGuardrail, BaseAWSLLM):
         request_data: dict | None,  # mutable-ok: proxy request body dict, mutated by the logging helper
         event_type: GuardrailEventHooks,
         start_time: "datetime",
+        completed_chunk_usages: list[BedrockGuardrailUsage],  # mutable-ok: passed through to the single-call layer
     ) -> BedrockGuardrailResponse:
         """Post one ApplyGuardrail call for `content`, retrying with exponential
         backoff on AWS ThrottlingException (HTTP 429).
@@ -1072,6 +1084,7 @@ class BedrockGuardrail(CustomGuardrail, BaseAWSLLM):
                     request_data=request_data,
                     event_type=event_type,
                     start_time=start_time,
+                    completed_chunk_usages=completed_chunk_usages,
                 )
             except HTTPException as exc:
                 if (
@@ -1093,6 +1106,7 @@ class BedrockGuardrail(CustomGuardrail, BaseAWSLLM):
         request_data: dict | None,  # mutable-ok: proxy request body dict, mutated by the logging helper
         event_type: GuardrailEventHooks,
         start_time: "datetime",
+        completed_chunk_usages: list[BedrockGuardrailUsage],  # mutable-ok: billed-chunk usage accumulator
     ) -> BedrockGuardrailResponse:
         """Make exactly one signed ApplyGuardrail HTTP call for `content` and
         parse the result. Raises HTTPException on a guardrail block or any
@@ -1108,7 +1122,10 @@ class BedrockGuardrail(CustomGuardrail, BaseAWSLLM):
 
         A block is logged here rather than by the caller: it ends the whole chunking
         flow immediately, with no further chunks attempted, so there is no later
-        merged response for the caller to log instead.
+        merged response for the caller to log instead. The logged usage still spans
+        the whole logical request: chunks that passed before the block appended what
+        AWS billed them to ``completed_chunk_usages``, and the attempt log sums those
+        with the blocking call's own usage.
         """
         bedrock_request_data: Final = {  # mutable-ok: outbound JSON request body
             **base_request_data,
@@ -1151,10 +1168,17 @@ class BedrockGuardrail(CustomGuardrail, BaseAWSLLM):
                     request_data=request_data,
                     event_type=event_type,
                     start_time=start_time,
+                    aws_region_name=aws_region_name,
+                    completed_chunk_usages=completed_chunk_usages,
                 )
                 raise self._get_http_exception_for_blocked_guardrail(
                     bedrock_guardrail_response, request_data=request_data
                 )
+            response_usage: Final = bedrock_guardrail_response.get("usage")
+            if isinstance(response_usage, dict):
+                completed_chunk_usages.append(
+                    response_usage
+                )  # rebind-ok: accumulator threaded from make_bedrock_api_request, recording this billed call
             return bedrock_guardrail_response
 
         status_code, detail_message = self._parse_bedrock_guardrail_error_response(httpx_response)
@@ -1172,14 +1196,31 @@ class BedrockGuardrail(CustomGuardrail, BaseAWSLLM):
         request_data: dict | None,  # mutable-ok: proxy request body dict, mutated by the logging helper
         event_type: GuardrailEventHooks,
         start_time: "datetime",
+        aws_region_name: str | None,
+        completed_chunk_usages: Sequence[BedrockGuardrailUsage],
     ) -> None:
-        """Log a single ApplyGuardrail HTTP attempt as-is (its own status,
-        derived from its own response). Used only for the blocked-content
-        case, which ends the whole chunking flow immediately."""
-        tracing_detail: Final = self._build_tracing_detail(BedrockGuardrailResponse(**json_response))
+        """Log the blocking ApplyGuardrail attempt, which ends the whole chunking
+        flow immediately. Its status derives from its own response, but its usage
+        (and so its cost) spans every billed call of the logical request: the
+        chunks that passed before the block plus the blocking call itself."""
+        blocking_usage: Final = json_response.get("usage")
+        billed_usages: Final[tuple[BedrockGuardrailUsage, ...]] = tuple(completed_chunk_usages) + (
+            (blocking_usage,) if isinstance(blocking_usage, dict) else ()
+        )
+        logged_json_response: Final = (
+            {  # mutable-ok: raw AWS JSON payload carrying the total billed usage
+                **json_response,
+                "usage": self._sum_usage_counters(billed_usages),
+            }
+            if completed_chunk_usages
+            else json_response
+        )
+        tracing_detail: Final = self._build_tracing_detail(
+            BedrockGuardrailResponse(**logged_json_response), aws_region_name=aws_region_name
+        )
         self.add_standard_logging_guardrail_information_to_request_data(
             guardrail_provider=self.guardrail_provider,
-            guardrail_json_response=json_response,
+            guardrail_json_response=logged_json_response,
             request_data=request_data or {},  # mutable-ok: logging helper requires a dict
             guardrail_status=self._get_bedrock_guardrail_response_status(response=httpx_response),
             start_time=start_time.timestamp(),
@@ -1195,6 +1236,7 @@ class BedrockGuardrail(CustomGuardrail, BaseAWSLLM):
         request_data: dict | None,  # mutable-ok: proxy request body dict, mutated by the logging helper
         event_type: GuardrailEventHooks,
         start_time: "datetime",
+        aws_region_name: str | None,
     ) -> None:
         """Log one logical ApplyGuardrail call -- possibly several chunk calls
         under the hood -- using its final merged response, so a chunked
@@ -1205,7 +1247,7 @@ class BedrockGuardrail(CustomGuardrail, BaseAWSLLM):
         ``Output.__type`` with an exception marker. That marker survives the merge,
         so the status is derived from the merged response rather than assumed to be
         a success, which is what the pre-chunking code reported for that shape."""
-        tracing_detail: Final = self._build_tracing_detail(merged_response)
+        tracing_detail: Final = self._build_tracing_detail(merged_response, aws_region_name=aws_region_name)
         self.add_standard_logging_guardrail_information_to_request_data(
             guardrail_provider=self.guardrail_provider,
             guardrail_json_response=dict(merged_response),  # mutable-ok: logging helper requires a dict
@@ -1228,20 +1270,36 @@ class BedrockGuardrail(CustomGuardrail, BaseAWSLLM):
         request_data: dict | None,  # mutable-ok: proxy request body dict, mutated by the logging helper
         event_type: GuardrailEventHooks,
         start_time: "datetime",
+        aws_region_name: str | None,
+        completed_chunk_usages: Sequence[BedrockGuardrailUsage],
     ) -> None:
         """Log one logical ApplyGuardrail call that failed end-to-end (an
         unrecoverable too-large error, a non-size validation error, or
         exhausted throttle retries) as a single failure, rather than logging
-        every failed attempt chunking made along the way."""
+        every failed attempt chunking made along the way. Chunk calls AWS
+        billed before the failure still carry their usage and cost."""
+        billed_usage: Final = self._sum_usage_counters(completed_chunk_usages) if completed_chunk_usages else None
+        error_payload: Final = {"error": str(detail)}  # mutable-ok: logging helper requires a dict
+        json_response: Final = (
+            {**error_payload, "usage": billed_usage}  # mutable-ok: logging helper requires a dict
+            if billed_usage is not None
+            else error_payload
+        )
+        tracing_detail: Final = (
+            self._build_tracing_detail(BedrockGuardrailResponse(usage=billed_usage), aws_region_name=aws_region_name)
+            if billed_usage is not None
+            else None
+        )
         self.add_standard_logging_guardrail_information_to_request_data(
             guardrail_provider=self.guardrail_provider,
-            guardrail_json_response={"error": str(detail)},  # mutable-ok: logging helper requires a dict
+            guardrail_json_response=json_response,
             request_data=request_data or {},  # mutable-ok: logging helper requires a dict
             guardrail_status="guardrail_failed_to_respond",
             start_time=start_time.timestamp(),
             end_time=datetime.now(timezone.utc).timestamp(),
             duration=(datetime.now(timezone.utc) - start_time).total_seconds(),
             event_type=event_type,
+            tracing_detail=tracing_detail or None,
         )
 
     @staticmethod
@@ -1504,15 +1562,20 @@ class BedrockGuardrail(CustomGuardrail, BaseAWSLLM):
         Keys are taken from the responses rather than from a fixed list, so a counter
         this code does not know about (AWS has added several) is still summed and
         reported instead of being silently dropped to zero."""
-        chunk_usages: Final = tuple(
-            chunk_result.response.get("usage") or {}  # mutable-ok: read-only empty fallback
-            for chunk_result in chunk_results
+        return BedrockGuardrail._sum_usage_counters(
+            tuple(
+                chunk_result.response.get("usage") or {}  # mutable-ok: read-only empty fallback
+                for chunk_result in chunk_results
+            )
         )
+
+    @staticmethod
+    def _sum_usage_counters(usages: Sequence[BedrockGuardrailUsage]) -> BedrockGuardrailUsage:
         return cast(  # cast-ok: TypedDict assembled from a comprehension
             BedrockGuardrailUsage,
             {  # mutable-ok: builds the TypedDict payload
-                key: sum(usage.get(key) or 0 for usage in chunk_usages)
-                for key in dict.fromkeys(key for usage in chunk_usages for key in usage)
+                key: sum(usage.get(key) or 0 for usage in usages)
+                for key in dict.fromkeys(key for usage in usages for key in usage)
             },
         )
 
@@ -2036,7 +2099,9 @@ class BedrockGuardrail(CustomGuardrail, BaseAWSLLM):
                 return (status_code, err)
         return (status_code, message)
 
-    def _build_tracing_detail(self, response: BedrockGuardrailResponse) -> GuardrailTracingDetail:
+    def _build_tracing_detail(
+        self, response: BedrockGuardrailResponse, aws_region_name: str | None
+    ) -> GuardrailTracingDetail:
         """
         Build the tracing detail from the raw Bedrock response, before
         redaction, so downstream loggers (OTEL, Langfuse, ...) get the
@@ -2060,6 +2125,9 @@ class BedrockGuardrail(CustomGuardrail, BaseAWSLLM):
             }
             if usage_units:
                 tracing_detail["guardrail_usage"] = usage_units
+                tracing_detail["guardrail_cost"] = bedrock_guardrail_cost(
+                    usage_units=usage_units, aws_region_name=aws_region_name
+                )
         return tracing_detail
 
     def _extract_violation_category_names(self, response: BedrockGuardrailResponse) -> list[str]:

litellm/proxy/hooks/proxy_track_cost_callback.py

@@ -11,6 +11,7 @@ from litellm.litellm_core_utils.core_helpers import (
     get_litellm_metadata_from_kwargs,
 )
 from litellm.litellm_core_utils.litellm_logging import StandardLoggingPayloadSetup
+from litellm.litellm_core_utils.llm_cost_calc.guardrail_cost import guardrail_information_cost
 from litellm.proxy._types import UserAPIKeyAuth
 from litellm.proxy.auth.auth_checks import (
     get_key_object,
@@ -184,9 +185,14 @@ class _ProxyDBLogger(CustomLogger):
         # recovered cost onto request_data (the usage rides along in
         # ``combined_usage_object`` for the token columns), so attribute the
         # real partial spend to this failure row instead of zero.
-        recovered_response_cost = 0.0
-        if isinstance(request_data.get("combined_usage_object"), litellm.Usage):
-            recovered_response_cost = max(float(request_data.get("response_cost") or 0.0), 0.0)
+        recovered_stream_cost: Final = (
+            max(float(request_data.get("response_cost") or 0.0), 0.0)
+            if isinstance(request_data.get("combined_usage_object"), litellm.Usage)
+            else 0.0
+        )
+        recovered_response_cost: Final = recovered_stream_cost + guardrail_information_cost(
+            existing_metadata.get("standard_logging_guardrail_information")
+        )
 
         await proxy_logging_obj.db_spend_update_writer.update_database(
             token=user_api_key_dict.api_key,

litellm/proxy/litellm_pre_call_utils.py

@@ -296,7 +296,10 @@ _ALLOW_CLIENT_MESSAGE_REDACTION_OPT_OUT_METADATA_KEY: Final = "allow_client_mess
 _CLIENT_PRICING_CONTROL_FIELDS: Final = frozenset(CustomPricingLiteLLMParams.model_fields.keys())
 # ``model_info`` carries the same pricing fields when read by
 # ``use_custom_pricing_for_model``; strip from metadata for the same reason.
-_CLIENT_PRICING_METADATA_FIELDS: Final = frozenset({"model_info"})
+# ``standard_logging_guardrail_information`` is proxy-written telemetry summed
+# into response_cost and spend; a client seeding it forges (even negative)
+# guardrail cost.
+_CLIENT_PRICING_METADATA_FIELDS: Final = frozenset({"model_info", "standard_logging_guardrail_information"})
 _ALLOW_CLIENT_PRICING_OVERRIDE_METADATA_KEY: Final = "allow_client_pricing_override"
 
 # Request fields whose value, when URL-valued, becomes the outbound destination

litellm/types/utils.py

@@ -196,6 +196,7 @@ class ModelInfoBase(ProviderSpecificModelInfo, total=False):
     input_cost_per_token: Required[float | None]
     input_cost_per_token_flex: float | None  # OpenAI flex service tier pricing
     input_cost_per_token_priority: float | None  # OpenAI priority service tier pricing
+    input_cost_per_token_ultrafast: ReadOnly[float | None]  # OpenAI ultrafast service tier pricing
     cache_creation_input_token_cost: float | None
     cache_creation_input_token_cost_above_200k_tokens: float | None
     cache_creation_input_token_cost_above_272k_tokens: float | None
@@ -204,9 +205,11 @@ class ModelInfoBase(ProviderSpecificModelInfo, total=False):
     cache_creation_input_token_cost_above_1hr: float | None
     cache_creation_input_token_cost_flex: float | None  # OpenAI flex service tier pricing
     cache_creation_input_token_cost_priority: float | None  # OpenAI priority service tier pricing
+    cache_creation_input_token_cost_ultrafast: ReadOnly[float | None]  # OpenAI ultrafast service tier pricing
     cache_read_input_token_cost: float | None
     cache_read_input_token_cost_flex: float | None  # OpenAI flex service tier pricing
     cache_read_input_token_cost_priority: float | None  # OpenAI priority service tier pricing
+    cache_read_input_token_cost_ultrafast: ReadOnly[float | None]  # OpenAI ultrafast service tier pricing
     cache_read_input_token_cost_above_200k_tokens: float | None
     cache_read_input_token_cost_above_200k_tokens_priority: float | None
     cache_read_input_token_cost_above_272k_tokens: float | None
@@ -238,6 +241,7 @@ class ModelInfoBase(ProviderSpecificModelInfo, total=False):
     output_cost_per_token: Required[float | None]
     output_cost_per_token_flex: float | None  # OpenAI flex service tier pricing
     output_cost_per_token_priority: float | None  # OpenAI priority service tier pricing
+    output_cost_per_token_ultrafast: ReadOnly[float | None]  # OpenAI ultrafast service tier pricing
     regional_processing_uplift_multiplier_eu: (
         float | None
     )  # OpenAI EU data-residency uplift multiplier applied to all token costs (e.g. 1.10 = +10%)
@@ -3023,6 +3027,11 @@ class StandardLoggingGuardrailInformation(TypedDict, total=False):
     provider's counter name (e.g. Bedrock's ``contentPolicyUnits``). Kept as a
     sibling of guardrail_response so spend-log prompt redaction never drops it."""
 
+    guardrail_cost: ReadOnly[float | None]
+    """USD cost of this guardrail invocation, priced from ``guardrail_usage`` by the
+    provider hook. Summed into the request's ``response_cost`` so it counts against
+    spend and budgets like token cost."""
+
 
 class EvalVerdict(TypedDict, total=False):
     criterion_name: str
@@ -3067,6 +3076,7 @@ class GuardrailTracingDetail(TypedDict, total=False):
     violation_categories: list[str] | None
     guardrail_action: str | None
     guardrail_usage: ReadOnly[Mapping[str, int] | None]
+    guardrail_cost: ReadOnly[float | None]
 
 
 StandardLoggingPayloadStatus = Literal["success", "failure"]
@@ -3106,8 +3116,9 @@ class CostBreakdown(TypedDict, total=False):
     cache_creation_cost: float  # Cost of cache-write tokens (premium rate)
     output_cost: float  # Cost of output/completion tokens (includes reasoning if applicable)
     reasoning_cost: float  # Cost of reasoning tokens (subset of output_cost)
-    total_cost: float  # Total cost (input + output + tool usage)
+    total_cost: ReadOnly[float]  # Total cost (input + output + tool usage + guardrail)
     tool_usage_cost: float  # Cost of usage of built-in tools
+    guardrail_cost: ReadOnly[float]  # Cost of guardrail invocations billed by the guardrail provider
     additional_costs: dict[str, float]  # Free-form additional costs (e.g., {"azure_model_router_flat_cost": 0.00014})
     original_cost: float  # Cost before discount (optional)
     discount_percent: float  # Discount percentage applied (e.g., 0.05 = 5%) (optional)
@@ -3294,6 +3305,7 @@ class CustomPricingLiteLLMParams(MirroredPricingParams):
     # This allows any model_info parameter to be set in litellm_params
     input_cost_per_token_flex: float | None = None
     input_cost_per_token_priority: float | None = None
+    input_cost_per_token_ultrafast: float | None = None
     cache_creation_input_token_cost_above_1hr: float | None = None
     cache_creation_input_token_cost_above_200k_tokens: float | None = None
     cache_creation_input_token_cost_above_272k_tokens: float | None = None
@@ -3301,9 +3313,11 @@ class CustomPricingLiteLLMParams(MirroredPricingParams):
     cache_creation_input_token_cost_above_272k_tokens_flex: float | None = None
     cache_creation_input_token_cost_flex: float | None = None
     cache_creation_input_token_cost_priority: float | None = None
+    cache_creation_input_token_cost_ultrafast: float | None = None
     cache_creation_input_audio_token_cost: float | None = None
     cache_read_input_token_cost_flex: float | None = None
     cache_read_input_token_cost_priority: float | None = None
+    cache_read_input_token_cost_ultrafast: float | None = None
     cache_read_input_token_cost_above_200k_tokens: float | None = None
     cache_read_input_token_cost_above_200k_tokens_priority: float | None = None
     cache_read_input_token_cost_above_272k_tokens_priority: float | None = None
@@ -3330,6 +3344,7 @@ class CustomPricingLiteLLMParams(MirroredPricingParams):
     output_cost_per_token_batches: float | None = None
     output_cost_per_token_flex: float | None = None
     output_cost_per_token_priority: float | None = None
+    output_cost_per_token_ultrafast: float | None = None
     output_cost_per_audio_token: float | None = None
     output_cost_per_token_above_128k_tokens: float | None = None
     output_cost_per_token_above_200k_tokens: float | None = None
@@ -4000,6 +4015,7 @@ class ServiceTier(Enum):
     FLEX = "flex"
     PRIORITY = "priority"
     FAST = "fast"
+    ULTRAFAST = "ultrafast"
 
 
 class DataResidency(Enum):

litellm/utils.py

@@ -5580,6 +5580,7 @@ def _get_model_info_helper(
                 input_cost_per_token=_input_cost_per_token,
                 input_cost_per_token_flex=_model_info.get("input_cost_per_token_flex", None),
                 input_cost_per_token_priority=_model_info.get("input_cost_per_token_priority", None),
+                input_cost_per_token_ultrafast=_model_info.get("input_cost_per_token_ultrafast", None),
                 cache_creation_input_token_cost=_model_info.get("cache_creation_input_token_cost", None),
                 cache_creation_input_token_cost_above_200k_tokens=_model_info.get(
                     "cache_creation_input_token_cost_above_200k_tokens", None
@@ -5597,6 +5598,9 @@ def _get_model_info_helper(
                 cache_creation_input_token_cost_priority=_model_info.get(
                     "cache_creation_input_token_cost_priority", None
                 ),
+                cache_creation_input_token_cost_ultrafast=_model_info.get(
+                    "cache_creation_input_token_cost_ultrafast", None
+                ),
                 cache_read_input_token_cost=_model_info.get("cache_read_input_token_cost", None),
                 prompt_cache_min_tokens=_model_info.get("prompt_cache_min_tokens", None),
                 cache_read_input_token_cost_above_200k_tokens=_model_info.get(
@@ -5619,6 +5623,7 @@ def _get_model_info_helper(
                 ),
                 cache_read_input_token_cost_flex=_model_info.get("cache_read_input_token_cost_flex", None),
                 cache_read_input_token_cost_priority=_model_info.get("cache_read_input_token_cost_priority", None),
+                cache_read_input_token_cost_ultrafast=_model_info.get("cache_read_input_token_cost_ultrafast", None),
                 cache_creation_input_token_cost_above_1hr=_model_info.get(
                     "cache_creation_input_token_cost_above_1hr", None
                 ),
@@ -5649,6 +5654,7 @@ def _get_model_info_helper(
                 output_cost_per_token=_output_cost_per_token,
                 output_cost_per_token_flex=_model_info.get("output_cost_per_token_flex", None),
                 output_cost_per_token_priority=_model_info.get("output_cost_per_token_priority", None),
+                output_cost_per_token_ultrafast=_model_info.get("output_cost_per_token_ultrafast", None),
                 regional_processing_uplift_multiplier_eu=_model_info.get(
                     "regional_processing_uplift_multiplier_eu", None
                 ),

model_prices_and_context_window.json

@@ -10052,6 +10052,21 @@
         "output_cost_per_second": 0.0066027,
         "supports_tool_choice": true
     },
+    "bedrock/guardrails": {
+        "guardrail_cost_per_unit": {
+            "automatedReasoningPolicyUnits": 0.00017,
+            "contentPolicyImageUnits": 0.00075,
+            "contentPolicyUnits": 0.00015,
+            "contextualGroundingPolicyUnits": 0.0001,
+            "sensitiveInformationPolicyFreeUnits": 0.0,
+            "sensitiveInformationPolicyUnits": 0.0001,
+            "topicPolicyUnits": 0.00015,
+            "wordPolicyUnits": 0.0
+        },
+        "litellm_provider": "bedrock",
+        "mode": "guardrail",
+        "source": "https://aws.amazon.com/bedrock/pricing/"
+    },
     "bedrock/ap-northeast-1/1-month-commitment/anthropic.claude-instant-v1": {
         "input_cost_per_second": 0.01475,
         "litellm_provider": "bedrock",

model_prices_and_context_window.schema.json

@@ -186,6 +186,14 @@
         "gemini_native_audio": {
           "type": "boolean"
         },
+        "guardrail_cost_per_unit": {
+          "type": "object",
+          "description": "USD cost per billable guardrail unit, keyed by the provider's usage counter name (e.g. Bedrock's contentPolicyUnits).",
+          "additionalProperties": {
+            "type": "number",
+            "minimum": 0
+          }
+        },
         "input_cost_per_audio_per_second": {
           "type": "number",
           "minimum": 0
@@ -361,6 +369,7 @@
             "chat",
             "completion",
             "embedding",
+            "guardrail",
             "image_edit",
             "image_generation",
             "moderation",

tests/test_litellm/litellm_core_utils/llm_cost_calc/test_guardrail_cost.py

@@ -0,0 +1,113 @@
+import os
+
+import pytest
+
+import litellm
+from litellm.litellm_core_utils.llm_cost_calc.guardrail_cost import (
+    bedrock_guardrail_cost,
+    cost_breakdown_with_guardrail,
+    guardrail_information_cost,
+)
+
+
+@pytest.fixture
+def synthetic_cost_map(monkeypatch):
+    monkeypatch.setattr(
+        litellm,
+        "model_cost",
+        {
+            "bedrock/guardrails": {
+                "guardrail_cost_per_unit": {
+                    "contentPolicyUnits": 0.00015,
+                    "topicPolicyUnits": 0.00015,
+                    "wordPolicyUnits": 0.0,
+                }
+            },
+            "bedrock/eu-west-1/guardrails": {"guardrail_cost_per_unit": {"contentPolicyUnits": 0.0002}},
+            "bedrock/us-west-2/guardrails": {"guardrail_cost_per_unit": "malformed"},
+        },
+    )
+
+
+def test_bedrock_guardrail_cost_prices_each_counter(synthetic_cost_map):
+    cost = bedrock_guardrail_cost(
+        usage_units={"contentPolicyUnits": 2, "topicPolicyUnits": 1, "wordPolicyUnits": 5},
+        aws_region_name="us-east-1",
+    )
+    assert cost == pytest.approx(0.00045)
+
+
+def test_bedrock_guardrail_cost_prefers_regional_entry(synthetic_cost_map):
+    cost = bedrock_guardrail_cost(usage_units={"contentPolicyUnits": 1}, aws_region_name="eu-west-1")
+    assert cost == pytest.approx(0.0002)
+
+
+def test_bedrock_guardrail_cost_unknown_counter_is_free(synthetic_cost_map):
+    assert bedrock_guardrail_cost(usage_units={"someFutureCounter": 3}, aws_region_name="us-east-1") == 0.0
+
+
+def test_bedrock_guardrail_cost_malformed_regional_entry_falls_back(synthetic_cost_map):
+    cost = bedrock_guardrail_cost(usage_units={"contentPolicyUnits": 1}, aws_region_name="us-west-2")
+    assert cost == pytest.approx(0.00015)
+
+
+def test_bedrock_guardrail_cost_no_pricing_entry(monkeypatch):
+    monkeypatch.setattr(litellm, "model_cost", {})
+    assert bedrock_guardrail_cost(usage_units={"contentPolicyUnits": 1}, aws_region_name="us-east-1") == 0.0
+
+
+def test_shipped_bedrock_guardrail_prices_match_aws_pricing_page():
+    os.environ["LITELLM_LOCAL_MODEL_COST_MAP"] = "True"
+    litellm.model_cost = litellm.get_model_cost_map(url="")
+    assert litellm.model_cost["bedrock/guardrails"]["guardrail_cost_per_unit"] == {
+        "automatedReasoningPolicyUnits": 0.00017,
+        "contentPolicyImageUnits": 0.00075,
+        "contentPolicyUnits": 0.00015,
+        "contextualGroundingPolicyUnits": 0.0001,
+        "sensitiveInformationPolicyFreeUnits": 0.0,
+        "sensitiveInformationPolicyUnits": 0.0001,
+        "topicPolicyUnits": 0.00015,
+        "wordPolicyUnits": 0.0,
+    }
+    assert "bedrock/guardrails" not in litellm.bedrock_models
+
+
+def test_guardrail_information_cost_sums_entries():
+    entries = [
+        {"guardrail_name": "a", "guardrail_cost": 0.0003},
+        {"guardrail_name": "b", "guardrail_cost": None},
+        {"guardrail_name": "c"},
+        {"guardrail_name": "d", "guardrail_cost": 0.0001},
+    ]
+    assert guardrail_information_cost(entries) == pytest.approx(0.0004)
+
+
+def test_guardrail_information_cost_single_entry_and_garbage():
+    assert guardrail_information_cost({"guardrail_cost": 0.0001}) == pytest.approx(0.0001)
+    assert guardrail_information_cost(None) == 0.0
+    assert guardrail_information_cost("not-guardrail-info") == 0.0
+    assert guardrail_information_cost([{"guardrail_cost": "bad"}]) == 0.0
+
+
+def test_guardrail_information_cost_ignores_negative_and_non_finite():
+    entries = [
+        {"guardrail_name": "forged-negative", "guardrail_cost": -0.005},
+        {"guardrail_name": "forged-nan", "guardrail_cost": float("nan")},
+        {"guardrail_name": "forged-inf", "guardrail_cost": float("inf")},
+        {"guardrail_name": "real", "guardrail_cost": 0.0003},
+    ]
+    assert guardrail_information_cost(entries) == pytest.approx(0.0003)
+    assert guardrail_information_cost({"guardrail_cost": -1.0}) == 0.0
+
+
+def test_cost_breakdown_with_guardrail_merges_and_creates():
+    assert cost_breakdown_with_guardrail(None, 0.0) is None
+    untouched = {"input_cost": 0.1, "total_cost": 0.4}
+    assert cost_breakdown_with_guardrail(untouched, 0.0) is untouched
+    merged = cost_breakdown_with_guardrail({"input_cost": 0.1, "total_cost": 0.4}, 0.0003)
+    assert merged is not None
+    assert merged["guardrail_cost"] == pytest.approx(0.0003)
+    assert merged["total_cost"] == pytest.approx(0.4003)
+    assert merged["input_cost"] == pytest.approx(0.1)
+    created = cost_breakdown_with_guardrail(None, 0.0003)
+    assert created == {"guardrail_cost": 0.0003, "total_cost": 0.0003}

tests/test_litellm/litellm_core_utils/llm_cost_calc/test_llm_cost_calc_utils.py

@@ -1781,6 +1781,81 @@ def test_service_tier_fallback_pricing():
     ), f"Standard completion cost mismatch: {std_cost[1]} vs {expected_standard_completion}"
 
 
+def test_service_tier_ultrafast_pricing():
+    """An ultrafast request bills the *_ultrafast rates for all token types.
+
+    Regression for the ultrafast service tier being absent from ServiceTier:
+    the cost-key lookup silently returned the standard keys, undercounting
+    every ultrafast request.
+    """
+    cached_tokens = 200
+    cache_write_tokens = 300
+    text_tokens = 500
+    usage = Usage(
+        prompt_tokens=text_tokens + cached_tokens + cache_write_tokens,
+        completion_tokens=400,
+        total_tokens=text_tokens + cached_tokens + cache_write_tokens + 400,
+        prompt_tokens_details=PromptTokensDetailsWrapper(
+            cached_tokens=cached_tokens, cache_write_tokens=cache_write_tokens
+        ),
+    )
+    model_info: ModelInfo = {
+        "key": "gpt-5.6-sol",
+        "input_cost_per_token": 5e-06,
+        "output_cost_per_token": 3e-05,
+        "cache_creation_input_token_cost": 6.25e-06,
+        "cache_read_input_token_cost": 5e-07,
+        "input_cost_per_token_ultrafast": 5e-05,
+        "output_cost_per_token_ultrafast": 3e-04,
+        "cache_creation_input_token_cost_ultrafast": 6.25e-05,
+        "cache_read_input_token_cost_ultrafast": 5e-06,
+    }
+
+    prompt_cost, completion_cost = generic_cost_per_token(
+        model="gpt-5.6-sol",
+        usage=usage,
+        custom_llm_provider="openai",
+        service_tier="ultrafast",
+        model_info=model_info,
+    )
+
+    expected_prompt_cost = (
+        text_tokens * 5e-05 + cached_tokens * 5e-06 + cache_write_tokens * 6.25e-05
+    )
+    assert prompt_cost == pytest.approx(expected_prompt_cost)
+    assert completion_cost == pytest.approx(400 * 3e-04)
+
+
+def test_service_tier_ultrafast_fallback_pricing():
+    """Without *_ultrafast keys an ultrafast request bills the standard rate, not zero.
+
+    Guards the suffix fallback in _get_cost_per_unit: "_fast" is a substring of
+    "_ultrafast", so a shortest-first suffix match would strip the wrong suffix
+    and price the request at 0.
+    """
+    os.environ["LITELLM_LOCAL_MODEL_COST_MAP"] = "True"
+    litellm.model_cost = litellm.get_model_cost_map(url="")
+
+    usage = Usage(prompt_tokens=1000, completion_tokens=500, total_tokens=1500)
+
+    std_prompt_cost, std_completion_cost = generic_cost_per_token(
+        model="gpt-5.6-sol",
+        usage=usage,
+        custom_llm_provider="openai",
+        service_tier=None,
+    )
+    ultrafast_prompt_cost, ultrafast_completion_cost = generic_cost_per_token(
+        model="gpt-5.6-sol",
+        usage=usage,
+        custom_llm_provider="openai",
+        service_tier="ultrafast",
+    )
+
+    assert std_prompt_cost + std_completion_cost > 0
+    assert ultrafast_prompt_cost == pytest.approx(std_prompt_cost)
+    assert ultrafast_completion_cost == pytest.approx(std_completion_cost)
+
+
 @pytest.mark.parametrize(
     "model",
     [
@@ -2322,7 +2397,11 @@ def test_service_tier_suffixes_constant_in_sync_with_enum():
     from litellm.litellm_core_utils.llm_cost_calc.utils import _SERVICE_TIER_SUFFIXES
     from litellm.types.utils import ServiceTier
 
-    assert _SERVICE_TIER_SUFFIXES == tuple(f"_{st.value}" for st in ServiceTier)
+    assert set(_SERVICE_TIER_SUFFIXES) == {f"_{st.value}" for st in ServiceTier}
+    # longest-first so a substring match resolves "_ultrafast" before "_fast"
+    assert list(_SERVICE_TIER_SUFFIXES) == sorted(
+        _SERVICE_TIER_SUFFIXES, key=len, reverse=True
+    )
 
 
 def test_get_cost_per_unit_falls_back_from_service_tier_key_to_base():

tests/test_litellm/litellm_core_utils/test_litellm_logging.py

@@ -4826,3 +4826,87 @@ async def test_restore_correlation_context_works_across_asyncio_task_boundary():
     finally:
         trace_id_var.set("")
         session_id_var.set("")
+
+
+def _build_success_payload(logging_obj, kwargs):
+    import datetime
+
+    from litellm.litellm_core_utils.litellm_logging import (
+        get_standard_logging_object_payload,
+    )
+
+    now = datetime.datetime.now()
+    return get_standard_logging_object_payload(
+        kwargs=kwargs,
+        init_response_obj={},
+        start_time=now,
+        end_time=now,
+        logging_obj=logging_obj,
+        status="success",
+    )
+
+
+def _guardrail_kwargs(response_cost):
+    return {
+        "litellm_call_id": "guardrail-cost-call",
+        "model": "gpt-4o",
+        "messages": [],
+        "response_cost": response_cost,
+        "litellm_params": {
+            "metadata": {
+                "standard_logging_guardrail_information": [
+                    {
+                        "guardrail_name": "bedrock-pre",
+                        "guardrail_status": "success",
+                        "guardrail_usage": {"topicPolicyUnits": 1, "contentPolicyUnits": 1},
+                        "guardrail_cost": 0.0003,
+                    },
+                    {"guardrail_name": "no-usage-guardrail", "guardrail_status": "success"},
+                ]
+            }
+        },
+    }
+
+
+def test_payload_response_cost_includes_guardrail_cost(logging_obj):
+    """LIT-5651: provider-billed guardrail cost must count in response_cost."""
+    payload = _build_success_payload(logging_obj, _guardrail_kwargs(response_cost=0.0000429))
+
+    assert payload is not None
+    assert payload["response_cost"] == pytest.approx(0.0003429)
+    assert payload["cost_breakdown"] is not None
+    assert payload["cost_breakdown"]["guardrail_cost"] == pytest.approx(0.0003)
+    assert payload["cost_breakdown"]["total_cost"] == pytest.approx(0.0003)
+    assert payload["hidden_params"]["response_cost"] == pytest.approx(0.0000429)
+
+
+def test_payload_guardrail_cost_merges_into_existing_cost_breakdown(logging_obj):
+    logging_obj.set_cost_breakdown(
+        input_cost=0.00003,
+        output_cost=0.0000129,
+        total_cost=0.0000429,
+        cost_for_built_in_tools_cost_usd_dollar=0.0,
+    )
+    payload = _build_success_payload(logging_obj, _guardrail_kwargs(response_cost=0.0000429))
+
+    assert payload is not None
+    assert payload["response_cost"] == pytest.approx(0.0003429)
+    assert payload["cost_breakdown"]["guardrail_cost"] == pytest.approx(0.0003)
+    assert payload["cost_breakdown"]["total_cost"] == pytest.approx(0.0003429)
+    assert payload["cost_breakdown"]["input_cost"] == pytest.approx(0.00003)
+    assert logging_obj.cost_breakdown["total_cost"] == pytest.approx(0.0000429)
+
+
+def test_payload_without_guardrail_cost_is_unchanged(logging_obj):
+    kwargs = {
+        "litellm_call_id": "no-guardrail-call",
+        "model": "gpt-4o",
+        "messages": [],
+        "response_cost": 0.0000429,
+        "litellm_params": {"metadata": {}},
+    }
+    payload = _build_success_payload(logging_obj, kwargs)
+
+    assert payload is not None
+    assert payload["response_cost"] == pytest.approx(0.0000429)
+    assert payload["cost_breakdown"] is None

tests/test_litellm/proxy/guardrails/guardrail_hooks/test_bedrock_guardrails.py

@@ -5079,24 +5079,200 @@ async def test_apply_guardrail_failure_logs_a_dict_not_a_bare_string():
     assert "error" in logged
 
 
-def test_build_tracing_detail_surfaces_usage_counters():
-    """LIT-5650: the billable usage block Bedrock returns per ApplyGuardrail call must
-    land on the tracing detail as guardrail_usage so it reaches spend logs as a
-    sibling of guardrail_response (which default redaction replaces wholesale)."""
+def test_build_tracing_detail_surfaces_usage_counters_and_cost(monkeypatch):
+    """LIT-5650/LIT-5651: AWS-billed usage must land as guardrail_usage priced into guardrail_cost."""
+    monkeypatch.setattr(
+        litellm,
+        "model_cost",
+        {
+            "bedrock/guardrails": {
+                "guardrail_cost_per_unit": {
+                    "topicPolicyUnits": 0.00015,
+                    "contentPolicyUnits": 0.00015,
+                    "wordPolicyUnits": 0.0,
+                }
+            }
+        },
+    )
     guardrail = BedrockGuardrail(guardrailIdentifier="test-guardrail", guardrailVersion="DRAFT")
 
     detail = guardrail._build_tracing_detail(
         {
             "action": "GUARDRAIL_INTERVENED",
             "usage": {"topicPolicyUnits": 1, "contentPolicyUnits": 2, "wordPolicyUnits": 0, "oddball": "not-an-int"},
-        }
+        },
+        aws_region_name="us-east-1",
     )
 
     assert detail["guardrail_usage"] == {"topicPolicyUnits": 1, "contentPolicyUnits": 2, "wordPolicyUnits": 0}
+    assert detail["guardrail_cost"] == pytest.approx(0.00045)
 
 
 def test_build_tracing_detail_omits_guardrail_usage_when_bedrock_reports_none():
     guardrail = BedrockGuardrail(guardrailIdentifier="test-guardrail", guardrailVersion="DRAFT")
 
-    assert "guardrail_usage" not in guardrail._build_tracing_detail({"action": "NONE"})
-    assert "guardrail_usage" not in guardrail._build_tracing_detail({"action": "NONE", "usage": {}})
+    for detail in (
+        guardrail._build_tracing_detail({"action": "NONE"}, aws_region_name="us-east-1"),
+        guardrail._build_tracing_detail({"action": "NONE", "usage": {}}, aws_region_name="us-east-1"),
+    ):
+        assert "guardrail_usage" not in detail
+        assert "guardrail_cost" not in detail
+
+
+@pytest.mark.asyncio
+async def test_blocked_chunk_logs_usage_and_cost_of_prior_passed_chunks(monkeypatch):
+    """LIT-5651 regression: a block on a later chunk must still bill the chunks AWS already processed."""
+    monkeypatch.setattr(
+        litellm,
+        "model_cost",
+        {
+            "bedrock/guardrails": {
+                "guardrail_cost_per_unit": {
+                    "contentPolicyUnits": 0.00015,
+                    "wordPolicyUnits": 0.0,
+                }
+            }
+        },
+    )
+    guardrail = BedrockGuardrail(
+        guardrailIdentifier="test-guardrail",
+        guardrailVersion="DRAFT",
+        chunk_budget_chars=40,
+    )
+
+    too_large_response = MagicMock()
+    too_large_response.status_code = 429
+    too_large_response.json.return_value = {
+        "message": "Input text size (60 text units) exceeds the maximum allowed (1 text units) for the content filter policy"
+    }
+
+    passed_chunk_response = MagicMock()
+    passed_chunk_response.status_code = 200
+    passed_chunk_response.json.return_value = {
+        "action": "NONE",
+        "outputs": [],
+        "assessments": [],
+        "usage": {"contentPolicyUnits": 2, "wordPolicyUnits": 1},
+    }
+
+    blocked_chunk_response = MagicMock()
+    blocked_chunk_response.status_code = 200
+    blocked_chunk_response.json.return_value = {
+        "action": "GUARDRAIL_INTERVENED",
+        "assessments": [{"contentPolicy": {"filters": [{"type": "HATE", "confidence": "HIGH", "action": "BLOCKED"}]}}],
+        "outputs": [{"text": "Content blocked"}],
+        "usage": {"contentPolicyUnits": 3},
+    }
+
+    mock_credentials = MagicMock()
+    mock_credentials.access_key = "test-access-key"
+    mock_credentials.secret_key = "test-secret-key"
+    mock_credentials.token = None
+
+    request_data = {
+        "model": "gpt-4o",
+        "messages": [
+            {"role": "user", "content": "a" * 30},
+            {"role": "user", "content": "b" * 30},
+        ],
+    }
+
+    with (
+        patch.object(guardrail.async_handler, "post", new_callable=AsyncMock) as mock_post,
+        patch.object(guardrail, "_load_credentials", return_value=(mock_credentials, "us-east-1")),
+        patch.object(guardrail, "_prepare_request", return_value=MagicMock()),
+    ):
+        mock_post.side_effect = [too_large_response, passed_chunk_response, blocked_chunk_response]
+
+        with pytest.raises(HTTPException):
+            await guardrail.make_bedrock_api_request(
+                source="INPUT",
+                messages=request_data["messages"],
+                request_data=request_data,
+            )
+
+    assert mock_post.call_count == 3
+    logged_entries = request_data["metadata"]["standard_logging_guardrail_information"]
+    assert len(logged_entries) == 1
+    logged = logged_entries[0]
+    assert logged["guardrail_usage"] == {"contentPolicyUnits": 5, "wordPolicyUnits": 1}
+    assert logged["guardrail_cost"] == pytest.approx(0.00075)
+    assert logged["guardrail_response"]["usage"] == {"contentPolicyUnits": 5, "wordPolicyUnits": 1}
+
+
+@pytest.mark.asyncio
+async def test_terminal_failure_logs_usage_and_cost_of_prior_passed_chunks(monkeypatch):
+    """LIT-5651 regression: a terminal failure on a later chunk must still bill the chunks AWS already processed."""
+    monkeypatch.setattr(
+        litellm,
+        "model_cost",
+        {
+            "bedrock/guardrails": {
+                "guardrail_cost_per_unit": {
+                    "contentPolicyUnits": 0.00015,
+                    "wordPolicyUnits": 0.0,
+                }
+            }
+        },
+    )
+    guardrail = BedrockGuardrail(
+        guardrailIdentifier="test-guardrail",
+        guardrailVersion="DRAFT",
+        chunk_budget_chars=40,
+    )
+
+    too_large_response = MagicMock()
+    too_large_response.status_code = 429
+    too_large_response.json.return_value = {
+        "message": "Input text size (60 text units) exceeds the maximum allowed (1 text units) for the content filter policy"
+    }
+
+    passed_chunk_response = MagicMock()
+    passed_chunk_response.status_code = 200
+    passed_chunk_response.json.return_value = {
+        "action": "NONE",
+        "outputs": [],
+        "assessments": [],
+        "usage": {"contentPolicyUnits": 2, "wordPolicyUnits": 1},
+    }
+
+    failed_chunk_response = MagicMock()
+    failed_chunk_response.status_code = 400
+    failed_chunk_response.json.return_value = {"message": "ValidationException: guardrail is in a failed state"}
+
+    mock_credentials = MagicMock()
+    mock_credentials.access_key = "test-access-key"
+    mock_credentials.secret_key = "test-secret-key"
+    mock_credentials.token = None
+
+    request_data = {
+        "model": "gpt-4o",
+        "messages": [
+            {"role": "user", "content": "a" * 30},
+            {"role": "user", "content": "b" * 30},
+        ],
+    }
+
+    with (
+        patch.object(guardrail.async_handler, "post", new_callable=AsyncMock) as mock_post,
+        patch.object(guardrail, "_load_credentials", return_value=(mock_credentials, "us-east-1")),
+        patch.object(guardrail, "_prepare_request", return_value=MagicMock()),
+    ):
+        mock_post.side_effect = [too_large_response, passed_chunk_response, failed_chunk_response]
+
+        with pytest.raises(HTTPException):
+            await guardrail.make_bedrock_api_request(
+                source="INPUT",
+                messages=request_data["messages"],
+                request_data=request_data,
+            )
+
+    assert mock_post.call_count == 3
+    logged_entries = request_data["metadata"]["standard_logging_guardrail_information"]
+    assert len(logged_entries) == 1
+    logged = logged_entries[0]
+    assert logged["guardrail_status"] == "guardrail_failed_to_respond"
+    assert logged["guardrail_usage"] == {"contentPolicyUnits": 2, "wordPolicyUnits": 1}
+    assert logged["guardrail_cost"] == pytest.approx(0.0003)
+    assert logged["guardrail_response"]["usage"] == {"contentPolicyUnits": 2, "wordPolicyUnits": 1}
+    assert "error" in logged["guardrail_response"]

tests/test_litellm/proxy/hooks/test_proxy_track_cost_callback.py

@@ -17,7 +17,7 @@ from litellm.proxy.hooks.proxy_track_cost_callback import (
     _should_track_cost_callback,
     _update_database_and_spend_counters,
 )
-from litellm.types.utils import CallTypes
+from litellm.types.utils import CallTypes, Usage
 
 
 @pytest.mark.asyncio
@@ -152,6 +152,70 @@ async def test_async_post_call_failure_hook_does_not_clobber_guardrail_info_in_m
         assert metadata["standard_logging_guardrail_information"] == metadata_bucket_info
 
 
+@pytest.mark.asyncio
+async def test_async_post_call_failure_hook_bills_guardrail_cost_on_blocked_request():
+    """LIT-5651: a request blocked by a guardrail never reaches the LLM, but the
+    guardrail invocation itself is billed by the provider. The failure row must
+    charge that cost against the key instead of recording zero spend."""
+    logger = _ProxyDBLogger()
+    request_data = {
+        "model": "gpt-4",
+        "messages": [{"role": "user", "content": "Hello"}],
+        "metadata": {
+            "standard_logging_guardrail_information": [
+                {
+                    "guardrail_name": "bedrock-guard",
+                    "guardrail_status": "guardrail_intervened",
+                    "guardrail_usage": {"topicPolicyUnits": 1, "contentPolicyUnits": 1},
+                    "guardrail_cost": 0.0003,
+                }
+            ]
+        },
+        "proxy_server_request": {"request_id": "test_request_id"},
+    }
+
+    with patch(
+        "litellm.proxy.db.db_spend_update_writer.DBSpendUpdateWriter.update_database",
+        new_callable=AsyncMock,
+    ) as mock_update_database:
+        await logger.async_post_call_failure_hook(
+            request_data=request_data,
+            original_exception=Exception("Violated guardrail policy"),
+            user_api_key_dict=UserAPIKeyAuth(api_key="test_api_key"),
+        )
+
+        assert mock_update_database.call_args[1]["response_cost"] == pytest.approx(0.0003)
+
+
+@pytest.mark.asyncio
+async def test_async_post_call_failure_hook_adds_guardrail_cost_to_recovered_stream_cost():
+    logger = _ProxyDBLogger()
+    request_data = {
+        "model": "gpt-4",
+        "messages": [{"role": "user", "content": "Hello"}],
+        "metadata": {
+            "standard_logging_guardrail_information": [
+                {"guardrail_name": "bedrock-guard", "guardrail_status": "success", "guardrail_cost": 0.0003}
+            ]
+        },
+        "proxy_server_request": {"request_id": "test_request_id"},
+        "combined_usage_object": Usage(prompt_tokens=10, completion_tokens=5, total_tokens=15),
+        "response_cost": 0.001,
+    }
+
+    with patch(
+        "litellm.proxy.db.db_spend_update_writer.DBSpendUpdateWriter.update_database",
+        new_callable=AsyncMock,
+    ) as mock_update_database:
+        await logger.async_post_call_failure_hook(
+            request_data=request_data,
+            original_exception=Exception("stream broke mid-flight"),
+            user_api_key_dict=UserAPIKeyAuth(api_key="test_api_key"),
+        )
+
+        assert mock_update_database.call_args[1]["response_cost"] == pytest.approx(0.0013)
+
+
 @pytest.mark.asyncio
 async def test_async_post_call_failure_hook_non_llm_route():
     # Setup

tests/test_litellm/proxy/test_pricing_field_strip.py

@@ -102,6 +102,30 @@ class TestStripClientPricingOverrides:
         assert data["metadata"] == {"user_session": "keep-me"}
         assert data["litellm_metadata"] == {}
 
+    def test_metadata_guardrail_information_dropped(self):
+        # Client-seeded guardrail entries would otherwise be summed into
+        # response_cost and spend, letting a caller forge (even negative)
+        # guardrail cost against their own budget.
+        data = {
+            "model": "gpt-4",
+            "metadata": {
+                "user_session": "keep-me",
+                "standard_logging_guardrail_information": [
+                    {
+                        "guardrail_name": "forged",
+                        "guardrail_status": "success",
+                        "guardrail_cost": -0.005,
+                    }
+                ],
+            },
+            "litellm_metadata": {
+                "standard_logging_guardrail_information": [{"guardrail_cost": 5.0}],
+            },
+        }
+        _strip_client_pricing_overrides(data)
+        assert data["metadata"] == {"user_session": "keep-me"}
+        assert data["litellm_metadata"] == {}
+
     def test_non_pricing_fields_untouched(self):
         data = {
             "model": "gpt-4",
@@ -129,6 +153,7 @@ class TestStripClientPricingOverrides:
 
     def test_metadata_field_set_contains_model_info(self):
         assert "model_info" in _CLIENT_PRICING_METADATA_FIELDS
+        assert "standard_logging_guardrail_information" in _CLIENT_PRICING_METADATA_FIELDS
 
     def test_strip_emits_debug_log_listing_dropped_fields(self, caplog):
         # Operators need a paper trail so they can diagnose why a previously

tests/test_litellm/test_utils.py

@@ -869,6 +869,7 @@ def test_aaamodel_prices_and_context_window_json_is_valid():
                         "container",
                         "image_edit",
                         "embedding",
+                        "guardrail",
                         "image_generation",
                         "video_generation",
                         "moderation",
@@ -976,6 +977,10 @@ def test_aaamodel_prices_and_context_window_json_is_valid():
                         "type": "string",
                     },
                 },
+                "guardrail_cost_per_unit": {
+                    "type": "object",
+                    "additionalProperties": {"type": "number"},
+                },
                 "search_context_cost_per_query": {
                     "type": "object",
                     "properties": {

ui/litellm-dashboard/eslint-suppressions.json

@@ -161,9 +161,6 @@
     "local/filename-pascal-case": {
       "count": 1
     },
-    "no-restricted-imports": {
-      "count": 2
-    },
     "react-hooks/set-state-in-effect": {
       "count": 1
     }
@@ -959,7 +956,7 @@
   },
   "src/app/(dashboard)/playground/components/compareUI/components/ModelSelector.tsx": {
     "no-restricted-imports": {
-      "count": 2
+      "count": 1
     }
   },
   "src/app/(dashboard)/playground/components/complianceUI/ComplianceUI.tsx": {
@@ -1790,7 +1787,7 @@
       "count": 2
     },
     "no-restricted-imports": {
-      "count": 2
+      "count": 1
     },
     "prefer-const": {
       "count": 2
@@ -2735,7 +2732,7 @@
   },
   "src/components/team/LoggingSettings.tsx": {
     "no-restricted-imports": {
-      "count": 2
+      "count": 1
     }
   },
   "src/components/team/TeamInfo.tsx": {
@@ -2746,7 +2743,7 @@
       "count": 3
     },
     "no-restricted-imports": {
-      "count": 2
+      "count": 1
     },
     "react-hooks/set-state-in-effect": {
       "count": 1

ui/litellm-dashboard/src/app/(dashboard)/caching/_components/cache_settings/index.integration.test.tsx

@@ -63,6 +63,19 @@ describe("CacheSettings advanced settings round-trip", () => {
     });
   });
 
+  it("reveals the advanced field sections only after the user expands them", async () => {
+    const user = userEvent.setup();
+    renderSettings();
+    await screen.findByText("Connection Settings");
+    expect(screen.queryByText("SSL Settings")).not.toBeInTheDocument();
+
+    await user.click(screen.getByRole("button", { name: "Advanced Settings" }));
+
+    expect(await screen.findByText("SSL Settings")).toBeInTheDocument();
+    expect(screen.getByText("Cache Management")).toBeInTheDocument();
+    expect(screen.getByText("GCP Authentication")).toBeInTheDocument();
+  });
+
   it("sends the same payload whether or not the advanced section was expanded", async () => {
     const user = userEvent.setup();
     renderSettings();

ui/litellm-dashboard/src/app/(dashboard)/caching/_components/cache_settings/index.tsx

@@ -1,7 +1,7 @@
 import React, { useState, useEffect, useCallback } from "react";
 import { ChevronRight } from "lucide-react";
 import { FormProvider, useForm } from "react-hook-form";
-import { Button } from "@tremor/react";
+import { Button } from "@/components/ui/button";
 import { Collapsible, CollapsibleContent, CollapsibleTrigger } from "@/components/ui/collapsible";
 import { getCacheSettingsCall, testCacheConnectionCall, updateCacheSettingsCall } from "@/components/networking";
 import { fetchAvailableModels, ModelGroup } from "@/components/llm_calls/fetch_models";

ui/litellm-dashboard/src/app/(dashboard)/cost-tracking/_components/add_margin_form.test.tsx

@@ -95,6 +95,33 @@ describe("AddMarginForm", () => {
     expect(onAddProvider).toHaveBeenCalledTimes(1);
   });
 
+  it("should report the edited percentage as the user types", async () => {
+    const onPercentageChange = vi.fn();
+    const user = userEvent.setup();
+    renderWithProviders(
+      <AddMarginForm {...DEFAULT_PROPS} percentageValue="1" onPercentageChange={onPercentageChange} />,
+    );
+
+    await user.type(screen.getByPlaceholderText("10"), "0");
+    expect(onPercentageChange).toHaveBeenCalledWith("10");
+  });
+
+  it("should report the edited fixed amount as the user types", async () => {
+    const onFixedAmountChange = vi.fn();
+    const user = userEvent.setup();
+    renderWithProviders(
+      <AddMarginForm
+        {...DEFAULT_PROPS}
+        marginType="fixed"
+        fixedAmountValue="0.00"
+        onFixedAmountChange={onFixedAmountChange}
+      />,
+    );
+
+    await user.type(screen.getByPlaceholderText("0.001"), "1");
+    expect(onFixedAmountChange).toHaveBeenCalledWith("0.001");
+  });
+
   it("should call onMarginTypeChange when the Fixed Amount radio is clicked", async () => {
     const onMarginTypeChange = vi.fn();
     const user = userEvent.setup();

ui/litellm-dashboard/src/app/(dashboard)/cost-tracking/_components/add_margin_form.tsx

@@ -172,7 +172,7 @@ const AddMarginForm: React.FC<AddMarginFormProps> = ({
 
         <div className="flex items-center justify-end space-x-3 pt-6 border-t border-border">
           <Button
-            type="button"
+            type="submit"
             onClick={onAddProvider}
             disabled={
               !selectedProvider ||

ui/litellm-dashboard/src/app/(dashboard)/cost-tracking/_components/add_provider_form.test.tsx

@@ -62,6 +62,15 @@ describe("AddProviderForm", () => {
     expect(onAddProvider).toHaveBeenCalledTimes(1);
   });
 
+  it("should report the edited discount as the user types", async () => {
+    const onDiscountChange = vi.fn();
+    const user = userEvent.setup();
+    renderWithProviders(<AddProviderForm {...DEFAULT_PROPS} newDiscount="1" onDiscountChange={onDiscountChange} />);
+
+    await user.type(screen.getByPlaceholderText("5"), "5");
+    expect(onDiscountChange).toHaveBeenCalledWith("15");
+  });
+
   it("should show the percent sign next to the discount input", () => {
     renderWithProviders(<AddProviderForm {...DEFAULT_PROPS} />);
     expect(screen.getByText("%")).toBeInTheDocument();

ui/litellm-dashboard/src/app/(dashboard)/cost-tracking/_components/cost_tracking_settings.integration.test.tsx

@@ -130,7 +130,7 @@ describe("CostTrackingSettings submit paths", () => {
     expect(stableDiscountCallbacks.handleAddProvider).toHaveBeenCalledWith("OpenAI", "5");
   });
 
-  it("leaves Enter inert in the margin field while the button still submits", async () => {
+  it("requests the margin exactly once when Enter is pressed in the percentage field", async () => {
     const user = userEvent.setup();
     renderWithProviders(<CostTrackingSettings {...ADMIN_PROPS} />);
     const header = screen.getByText("Fee/Price Margin").closest("button");
@@ -142,14 +142,7 @@ describe("CostTrackingSettings submit paths", () => {
     await user.click((await screen.findAllByRole("option"))[0]);
     await user.type(screen.getByLabelText(/Margin Percentage/i), "10{Enter}");
 
-    expect(stableMarginCallbacks.handleAddMargin).not.toHaveBeenCalled();
-
-    const submit = screen
-      .getAllByRole("button")
-      .filter((button) => (button.textContent || "").trim() === "Add Provider Margin")
-      .pop()!;
-    await user.click(submit);
-
-    await waitFor(() => expect(stableMarginCallbacks.handleAddMargin).toHaveBeenCalledTimes(1));
+    await waitFor(() => expect(stableMarginCallbacks.handleAddMargin).toHaveBeenCalled());
+    expect(stableMarginCallbacks.handleAddMargin).toHaveBeenCalledTimes(1);
   });
 });

ui/litellm-dashboard/src/app/(dashboard)/cost-tracking/_components/cost_tracking_settings.tsx

@@ -382,7 +382,7 @@ const CostTrackingSettings: React.FC<CostTrackingSettingsProps> = ({ userID, use
             Select a provider (or &quot;Global&quot; for all providers) and configure the margin. You can use
             percentage-based or fixed amount.
           </p>
-          <div className="space-y-6">
+          <form onSubmit={(event) => event.preventDefault()} className="space-y-6">
             <AddMarginForm
               marginConfig={marginConfig}
               selectedProvider={selectedMarginProvider}
@@ -395,7 +395,7 @@ const CostTrackingSettings: React.FC<CostTrackingSettingsProps> = ({ userID, use
               onFixedAmountChange={setFixedAmountValue}
               onAddProvider={handleAddMargin}
             />
-          </div>
+          </form>
         </div>
       </Modal>
     </div>

ui/litellm-dashboard/src/app/(dashboard)/playground/components/compareUI/components/ModelSelector.test.tsx

@@ -1,4 +1,4 @@
-import { render, screen } from "@testing-library/react";
+import { fireEvent, render, screen } from "@testing-library/react";
 import userEvent from "@testing-library/user-event";
 import { describe, expect, it, vi } from "vitest";
 import { ModelSelector } from "./ModelSelector";
@@ -28,6 +28,18 @@ describe("ModelSelector", () => {
     expect(screen.getByTitle("custom-model-123")).toHaveTextContent("custom-model-123");
   });
 
+  it("reports a custom model typed into the custom name field", async () => {
+    const user = userEvent.setup();
+    const onChange = vi.fn();
+    render(<ModelSelector value="" onChange={onChange} models={MODELS} />);
+
+    await user.click(screen.getByRole("combobox"));
+    fireEvent.click(await screen.findByTitle("+ Add custom model"));
+    await user.type(await screen.findByPlaceholderText("Custom Model Name (Enter to add)"), "my-custom-model{Enter}");
+
+    expect(onChange).toHaveBeenCalledWith("my-custom-model");
+  });
+
   it("disables the control when disabled is set", () => {
     const { rerender } = render(<ModelSelector value="custom-model-123" onChange={vi.fn()} models={MODELS} />);
     expect(screen.getByRole("combobox")).toBeEnabled();

ui/litellm-dashboard/src/app/(dashboard)/playground/components/compareUI/components/ModelSelector.tsx

@@ -1,6 +1,6 @@
 import React, { useMemo, useState } from "react";
 import { Select } from "antd";
-import { TextInput } from "@tremor/react";
+import { Input } from "@/components/ui/input";
 interface ModelSelectorProps {
   value: string;
   onChange: (value: string) => void;
@@ -68,11 +68,11 @@ export function ModelSelector({ value, onChange, models, loading, disabled }: Mo
         <Select.Option value="__custom__">+ Add custom model</Select.Option>
       </Select>
       {isAddingCustom && (
-        <TextInput
+        <Input
           className="mt-2"
           placeholder="Custom Model Name (Enter to add)"
           value={customValue}
-          onValueChange={setCustomValue}
+          onChange={(e) => setCustomValue(e.target.value)}
           onKeyDown={(event) => {
             if (event.key === "Enter") {
               event.preventDefault();

ui/litellm-dashboard/src/components/Teams.tsx

@@ -5,9 +5,10 @@ import TeamInfoView from "@/components/team/TeamInfo";
 import TeamSSOSettings from "@/components/TeamSSOSettings";
 import { isProxyAdminRole } from "@/utils/roles";
 import { InfoCircleOutlined } from "@ant-design/icons";
-import { Accordion, AccordionBody, AccordionHeader, TextInput } from "@tremor/react";
+import { Collapsible, CollapsibleContent, CollapsibleTrigger } from "@/components/ui/collapsible";
+import { Input as UIInput } from "@/components/ui/input";
 import { Button, Form, Input, Layout, Modal, Select, Switch, Tabs, theme, Tooltip, Typography } from "antd";
-import { Plus, Users } from "lucide-react";
+import { ChevronDown, Plus, Users } from "lucide-react";
 import React, { useEffect, useState } from "react";
 import { useQuery, useQueryClient } from "@tanstack/react-query";
 import { PageHeader } from "@/components/shared/PageHeader";
@@ -542,7 +543,7 @@ const Teams: React.FC<TeamProps> = ({ accessToken, userID, userRole, premiumUser
                   },
                 ]}
               >
-                <TextInput placeholder="" data-testid="team-name-input" />
+                <UIInput data-testid="team-name-input" />
               </Form.Item>
               {(() => {
                 const adminOrgs = getAdminOrganizations(userRole, userID, organizations);
@@ -684,17 +685,18 @@ const Teams: React.FC<TeamProps> = ({ accessToken, userID, userRole, premiumUser
                 />
               </Form.Item>
 
-              <Accordion className="mt-20 mb-8">
-                <AccordionHeader>
+              <Collapsible className="mt-20 mb-8 overflow-hidden rounded-lg border">
+                <CollapsibleTrigger className="group/section flex w-full items-center justify-between px-4 py-3 text-left">
                   <b>Additional Settings</b>
-                </AccordionHeader>
-                <AccordionBody>
+                  <ChevronDown className="size-5 shrink-0 text-gray-500 transition-transform group-data-[panel-open]/section:rotate-180" />
+                </CollapsibleTrigger>
+                <CollapsibleContent className="px-4 pb-3">
                   <Form.Item
                     label="Team ID"
                     name="team_id"
                     help="ID of the team you want to create. If not provided, it will be generated automatically."
                   >
-                    <TextInput
+                    <UIInput
                       onChange={(e) => {
                         e.target.value = e.target.value.trim();
                       }}
@@ -713,7 +715,7 @@ const Teams: React.FC<TeamProps> = ({ accessToken, userID, userRole, premiumUser
                     name="team_member_key_duration"
                     tooltip="Set a limit to the duration of a team member's key. Format: 30s (seconds), 30m (minutes), 30h (hours), 30d (days), 1mo (month)"
                   >
-                    <TextInput placeholder="e.g., 30d" />
+                    <UIInput placeholder="e.g., 30d" />
                   </Form.Item>
                   <Form.Item
                     label="Team Member RPM Limit"
@@ -898,14 +900,15 @@ const Teams: React.FC<TeamProps> = ({ accessToken, userID, userRole, premiumUser
                       disabled={!premiumUser || !isProxyAdminRole(userRole || "")}
                     />
                   </Form.Item>
-                </AccordionBody>
-              </Accordion>
+                </CollapsibleContent>
+              </Collapsible>
 
-              <Accordion className="mt-8 mb-8">
-                <AccordionHeader>
+              <Collapsible className="mt-8 mb-8 overflow-hidden rounded-lg border">
+                <CollapsibleTrigger className="group/section flex w-full items-center justify-between px-4 py-3 text-left">
                   <b>MCP Settings</b>
-                </AccordionHeader>
-                <AccordionBody>
+                  <ChevronDown className="size-5 shrink-0 text-gray-500 transition-transform group-data-[panel-open]/section:rotate-180" />
+                </CollapsibleTrigger>
+                <CollapsibleContent className="px-4 pb-3">
                   <Form.Item
                     label={
                       <span>
@@ -951,14 +954,15 @@ const Teams: React.FC<TeamProps> = ({ accessToken, userID, userRole, premiumUser
                       </div>
                     )}
                   </Form.Item>
-                </AccordionBody>
-              </Accordion>
+                </CollapsibleContent>
+              </Collapsible>
 
-              <Accordion className="mt-8 mb-8">
-                <AccordionHeader>
+              <Collapsible className="mt-8 mb-8 overflow-hidden rounded-lg border">
+                <CollapsibleTrigger className="group/section flex w-full items-center justify-between px-4 py-3 text-left">
                   <b>Agent Settings</b>
-                </AccordionHeader>
-                <AccordionBody>
+                  <ChevronDown className="size-5 shrink-0 text-gray-500 transition-transform group-data-[panel-open]/section:rotate-180" />
+                </CollapsibleTrigger>
+                <CollapsibleContent className="px-4 pb-3">
                   <Form.Item
                     label={
                       <span>
@@ -979,14 +983,15 @@ const Teams: React.FC<TeamProps> = ({ accessToken, userID, userRole, premiumUser
                       placeholder="Select agents or access groups (optional)"
                     />
                   </Form.Item>
-                </AccordionBody>
-              </Accordion>
+                </CollapsibleContent>
+              </Collapsible>
 
-              <Accordion className="mt-8 mb-8">
-                <AccordionHeader>
+              <Collapsible className="mt-8 mb-8 overflow-hidden rounded-lg border">
+                <CollapsibleTrigger className="group/section flex w-full items-center justify-between px-4 py-3 text-left">
                   <b>Search Tool Settings</b>
-                </AccordionHeader>
-                <AccordionBody>
+                  <ChevronDown className="size-5 shrink-0 text-gray-500 transition-transform group-data-[panel-open]/section:rotate-180" />
+                </CollapsibleTrigger>
+                <CollapsibleContent className="px-4 pb-3">
                   <Form.Item
                     label={
                       <span>
@@ -1007,14 +1012,15 @@ const Teams: React.FC<TeamProps> = ({ accessToken, userID, userRole, premiumUser
                       placeholder="Select search tools (optional, empty = all allowed)"
                     />
                   </Form.Item>
-                </AccordionBody>
-              </Accordion>
+                </CollapsibleContent>
+              </Collapsible>
 
-              <Accordion className="mt-8 mb-8">
-                <AccordionHeader>
+              <Collapsible className="mt-8 mb-8 overflow-hidden rounded-lg border">
+                <CollapsibleTrigger className="group/section flex w-full items-center justify-between px-4 py-3 text-left">
                   <b>Logging Settings</b>
-                </AccordionHeader>
-                <AccordionBody>
+                  <ChevronDown className="size-5 shrink-0 text-gray-500 transition-transform group-data-[panel-open]/section:rotate-180" />
+                </CollapsibleTrigger>
+                <CollapsibleContent className="px-4 pb-3">
                   <div className="mt-4">
                     <PremiumLoggingSettings
                       value={loggingSettings}
@@ -1022,14 +1028,18 @@ const Teams: React.FC<TeamProps> = ({ accessToken, userID, userRole, premiumUser
                       premiumUser={premiumUser}
                     />
                   </div>
-                </AccordionBody>
-              </Accordion>
+                </CollapsibleContent>
+              </Collapsible>
 
-              <Accordion key={`router-settings-accordion-${routerSettingsKey}`} className="mt-8 mb-8">
-                <AccordionHeader>
+              <Collapsible
+                key={`router-settings-accordion-${routerSettingsKey}`}
+                className="mt-8 mb-8 overflow-hidden rounded-lg border"
+              >
+                <CollapsibleTrigger className="group/section flex w-full items-center justify-between px-4 py-3 text-left">
                   <b>Router Settings</b>
-                </AccordionHeader>
-                <AccordionBody>
+                  <ChevronDown className="size-5 shrink-0 text-gray-500 transition-transform group-data-[panel-open]/section:rotate-180" />
+                </CollapsibleTrigger>
+                <CollapsibleContent className="px-4 pb-3">
                   <div className="mt-4 w-full">
                     <RouterSettingsAccordion
                       key={routerSettingsKey}
@@ -1041,14 +1051,15 @@ const Teams: React.FC<TeamProps> = ({ accessToken, userID, userRole, premiumUser
                       }
                     />
                   </div>
-                </AccordionBody>
-              </Accordion>
+                </CollapsibleContent>
+              </Collapsible>
 
-              <Accordion className="mt-8 mb-8">
-                <AccordionHeader>
+              <Collapsible className="mt-8 mb-8 overflow-hidden rounded-lg border">
+                <CollapsibleTrigger className="group/section flex w-full items-center justify-between px-4 py-3 text-left">
                   <b>Model Aliases</b>
-                </AccordionHeader>
-                <AccordionBody>
+                  <ChevronDown className="size-5 shrink-0 text-gray-500 transition-transform group-data-[panel-open]/section:rotate-180" />
+                </CollapsibleTrigger>
+                <CollapsibleContent className="px-4 pb-3">
                   <div className="mt-4">
                     <Text type="secondary" style={{ fontSize: 14, marginBottom: 16, display: "block" }}>
                       Create custom aliases for models that can be used by team members in API calls. This allows you to
@@ -1061,8 +1072,8 @@ const Teams: React.FC<TeamProps> = ({ accessToken, userID, userRole, premiumUser
                       showExampleConfig={false}
                     />
                   </div>
-                </AccordionBody>
-              </Accordion>
+                </CollapsibleContent>
+              </Collapsible>
             </>
             <div style={{ textAlign: "right", marginTop: "10px" }}>
               <Button htmlType="submit" data-testid="create-team-submit">

ui/litellm-dashboard/src/components/team/LoggingSettings.test.tsx

@@ -1,5 +1,6 @@
 import React from "react";
 import { describe, it, expect, vi, beforeEach } from "vitest";
+import userEvent from "@testing-library/user-event";
 import { renderWithProviders, screen, fireEvent } from "../../../tests/test-utils";
 import LoggingSettings from "./LoggingSettings";
 
@@ -109,6 +110,29 @@ describe("LoggingSettings", () => {
     expect(updatedConfig[0].callback_vars.langsmith_sampling_rate).toBe("0.3"); // Preserves initial value
   });
 
+  it("masks a sensitive parameter until the reveal toggle is used", async () => {
+    const user = userEvent.setup({ delay: null });
+    const initialValue = [
+      {
+        callback_name: "langsmith",
+        callback_type: "success",
+        callback_vars: { langsmith_api_key: "sk-secret-value" },
+      },
+    ];
+
+    renderWithProviders(<LoggingSettings value={initialValue} onChange={vi.fn()} />);
+
+    const apiKeyInput = screen.getByPlaceholderText("os.environ/LANGSMITH_API_KEY");
+    expect(apiKeyInput).toHaveAttribute("type", "password");
+
+    await user.click(screen.getByRole("button", { name: "Show password" }));
+    expect(apiKeyInput).toHaveAttribute("type", "text");
+    expect(apiKeyInput).toHaveValue("sk-secret-value");
+
+    await user.click(screen.getByRole("button", { name: "Hide password" }));
+    expect(apiKeyInput).toHaveAttribute("type", "password");
+  });
+
   it("shows the bundled logo in the integration card header", () => {
     const initialValue = [
       {

ui/litellm-dashboard/src/components/team/LoggingSettings.tsx

@@ -2,14 +2,51 @@
 import React from "react";
 import { Select, Tooltip, Divider } from "antd";
 import { InfoCircleOutlined } from "@ant-design/icons";
-import { Button, Card, TextInput } from "@tremor/react";
-import { PlusIcon, TrashIcon, CogIcon, BanIcon } from "@heroicons/react/outline";
+import { Button } from "@/components/ui/button";
+import { Card } from "@/components/ui/card";
+import { Input } from "@/components/ui/input";
+import { InputGroup, InputGroupAddon, InputGroupButton, InputGroupInput } from "@/components/ui/input-group";
+import { CogIcon, BanIcon } from "@heroicons/react/outline";
+import { Eye, EyeOff, Plus, Trash2 } from "lucide-react";
 import { callbackInfo, callback_map, mapDisplayToInternalNames } from "../callback_info_helpers";
 import { Logo } from "@/components/molecules/logo/Logo";
 import NumericalInput from "../shared/numerical_input";
 
 const { Option } = Select;
 
+const CallbackVarInput: React.FC<{
+  sensitive: boolean;
+  placeholder: string;
+  value: string;
+  onValueChange: (value: string) => void;
+}> = ({ sensitive, placeholder, value, onValueChange }) => {
+  const [revealed, setRevealed] = React.useState(false);
+
+  if (!sensitive) {
+    return <Input placeholder={placeholder} value={value} onChange={(e) => onValueChange(e.target.value)} />;
+  }
+
+  return (
+    <InputGroup>
+      <InputGroupInput
+        type={revealed ? "text" : "password"}
+        placeholder={placeholder}
+        value={value}
+        onChange={(e) => onValueChange(e.target.value)}
+      />
+      <InputGroupAddon align="inline-end">
+        <InputGroupButton
+          size="icon-xs"
+          onClick={() => setRevealed(!revealed)}
+          aria-label={revealed ? "Hide password" : "Show password"}
+        >
+          {revealed ? <EyeOff /> : <Eye />}
+        </InputGroupButton>
+      </InputGroupAddon>
+    </InputGroup>
+  );
+};
+
 interface LoggingConfig {
   callback_name: string;
   callback_type: string;
@@ -138,11 +175,11 @@ const LoggingSettings: React.FC<LoggingSettingsProps> = ({
                   onChange={(e: any) => updateCallbackVar(configIndex, paramName, e.target.value)}
                 />
               ) : (
-                <TextInput
-                  type={paramType === "password" ? "password" : "text"}
+                <CallbackVarInput
+                  sensitive={paramType === "password"}
                   placeholder={`os.environ/${paramName.toUpperCase()}`}
                   value={config.callback_vars[paramName] || ""}
-                  onChange={(e) => updateCallbackVar(configIndex, paramName, e.target.value)}
+                  onValueChange={(newValue) => updateCallbackVar(configIndex, paramName, newValue)}
                 />
               )}
             </div>
@@ -212,11 +249,11 @@ const LoggingSettings: React.FC<LoggingSettingsProps> = ({
         <Button
           variant="secondary"
           onClick={addLoggingConfig}
-          icon={PlusIcon}
           size="sm"
           className="hover:border-blue-400 hover:text-blue-500"
           type="button"
         >
+          <Plus />
           Add Integration
         </Button>
       </div>
@@ -230,9 +267,7 @@ const LoggingSettings: React.FC<LoggingSettingsProps> = ({
           return (
             <Card
               key={index}
-              className="border border-gray-200 shadow-xs hover:shadow-md transition-shadow duration-200"
-              decoration="top"
-              decorationColor="blue"
+              className="block p-6 border border-gray-200 shadow-xs hover:shadow-md transition-shadow duration-200"
             >
               <div className="flex justify-between items-start mb-4">
                 <div className="flex items-center space-x-2">
@@ -246,14 +281,13 @@ const LoggingSettings: React.FC<LoggingSettingsProps> = ({
                   <span className="text-sm font-medium">{callbackDisplayName || "New Integration"} Configuration</span>
                 </div>
                 <Button
-                  variant="light"
+                  variant="ghost"
                   onClick={() => removeLoggingConfig(index)}
-                  icon={TrashIcon}
-                  size="xs"
-                  color="red"
-                  className="hover:bg-red-50"
+                  size="sm"
+                  className="text-red-500 hover:bg-red-50 hover:text-red-500"
                   type="button"
                 >
+                  <Trash2 />
                   Remove
                 </Button>
               </div>

ui/litellm-dashboard/src/components/team/TeamInfo.tsx

@@ -29,10 +29,14 @@ import {
   SaveOutlined,
 } from "@ant-design/icons";
 import { ArrowLeftIcon } from "@heroicons/react/outline";
-import { Accordion, AccordionBody, AccordionHeader, Badge, Card, Grid, Text, TextInput, Title } from "@tremor/react";
+import { StatusBadge, type StatusTone } from "@/components/shared/table_cells/status_badge";
+import { Badge } from "@/components/ui/badge";
+import { Card } from "@/components/ui/card";
+import { Collapsible, CollapsibleContent, CollapsibleTrigger } from "@/components/ui/collapsible";
+import { Input as UIInput } from "@/components/ui/input";
 import { Button, Form, Input, InputNumber, Select, Space, Switch, Tabs, Tag, Tooltip } from "antd";
 import { toast } from "@/lib/toast";
-import { CheckIcon, CopyIcon } from "lucide-react";
+import { CheckIcon, ChevronDown, CopyIcon } from "lucide-react";
 import React, { useEffect, useMemo, useState } from "react";
 import { copyToClipboard as utilCopyToClipboard } from "../../utils/dataUtils";
 import AccessGroupSelector from "../common_components/AccessGroupSelector";
@@ -92,11 +96,11 @@ const UI_MANAGED_METADATA_KEYS: ReadonlySet<string> = new Set([
   "disable_global_guardrails",
 ]);
 
-const TEAM_MODEL_BADGE_COLORS: Record<TeamModelBadgeKind, "red" | "gray" | "blue" | "green"> = {
-  "all-proxy": "red",
-  "no-default": "gray",
-  direct: "blue",
-  "access-group": "green",
+const TEAM_MODEL_BADGE_TONES: Record<TeamModelBadgeKind, StatusTone> = {
+  "all-proxy": "error",
+  "no-default": "neutral",
+  direct: "info",
+  "access-group": "success",
 };
 
 export interface TeamMembership {
@@ -732,9 +736,9 @@ const TeamInfoView: React.FC<TeamInfoProps> = ({
           <Button type="text" icon={<ArrowLeftIcon className="h-4 w-4" />} onClick={onClose} className="mb-4">
             Back to Teams
           </Button>
-          <Title>{info.team_alias}</Title>
+          <h1 className="text-2xl font-semibold">{info.team_alias}</h1>
           <div className="flex items-center">
-            <Text className="text-gray-500 font-mono">{info.team_id}</Text>
+            <p className="text-sm text-gray-500 font-mono">{info.team_id}</p>
             <Button
               type="text"
               size="small"
@@ -758,30 +762,30 @@ const TeamInfoView: React.FC<TeamInfoProps> = ({
             key: TEAM_INFO_TAB_KEYS.OVERVIEW,
             label: TEAM_INFO_TAB_LABELS[TEAM_INFO_TAB_KEYS.OVERVIEW],
             children: (
-              <Grid numItems={1} numItemsSm={2} numItemsLg={3} className="gap-6">
-                <Card>
-                  <Text>Budget Status</Text>
+              <div className="grid grid-cols-1 sm:grid-cols-2 lg:grid-cols-3 gap-6">
+                <Card className="block p-6">
+                  <p>Budget Status</p>
                   <div className="mt-2">
-                    <Title>${formatNumberWithCommas(info.spend, 2)}</Title>
-                    <Text>
+                    <h3 className="text-lg font-medium">${formatNumberWithCommas(info.spend, 2)}</h3>
+                    <p>
                       of {info.max_budget === null ? "Unlimited" : `$${formatNumberWithCommas(info.max_budget, 2)}`}
-                    </Text>
-                    {info.budget_duration && <Text className="text-gray-500">Reset: {info.budget_duration}</Text>}
+                    </p>
+                    {info.budget_duration && <p className="text-gray-500">Reset: {info.budget_duration}</p>}
                     <br />
                     {info.team_member_budget_table && (
-                      <Text className="text-gray-500">
+                      <p className="text-gray-500">
                         Team Member Budget: ${formatNumberWithCommas(info.team_member_budget_table.max_budget, 2)}
-                      </Text>
+                      </p>
                     )}
                   </div>
                 </Card>
 
-                <Card>
-                  <Text>Rate Limits</Text>
+                <Card className="block p-6">
+                  <p>Rate Limits</p>
                   <div className="mt-2">
-                    <Text>TPM: {info.tpm_limit || "Unlimited"}</Text>
-                    <Text>RPM: {info.rpm_limit || "Unlimited"}</Text>
-                    {info.max_parallel_requests && <Text>Max Parallel Requests: {info.max_parallel_requests}</Text>}
+                    <p>TPM: {info.tpm_limit || "Unlimited"}</p>
+                    <p>RPM: {info.rpm_limit || "Unlimited"}</p>
+                    {info.max_parallel_requests && <p>Max Parallel Requests: {info.max_parallel_requests}</p>}
                     {(() => {
                       const modelTpm = (info.metadata?.model_tpm_limit ?? {}) as Record<string, number>;
                       const modelRpm = (info.metadata?.model_rpm_limit ?? {}) as Record<string, number>;
@@ -789,33 +793,33 @@ const TeamInfoView: React.FC<TeamInfoProps> = ({
                       if (models.length === 0) return null;
                       return (
                         <div className="mt-3">
-                          <Text className="text-gray-500">Per-model limits:</Text>
+                          <p className="text-gray-500">Per-model limits:</p>
                           {models.map((m) => (
-                            <Text key={m} className="text-xs">
+                            <p key={m} className="text-xs">
                               {m}: TPM {modelTpm[m] ?? "—"}, RPM {modelRpm[m] ?? "—"}
-                            </Text>
+                            </p>
                           ))}
                         </div>
                       );
                     })()}
-                    <Text>Estimated Output Tokens: {info.metadata?.default_estimated_output_tokens ?? "Default"}</Text>
-                    <Text>
+                    <p>Estimated Output Tokens: {info.metadata?.default_estimated_output_tokens ?? "Default"}</p>
+                    <p>
                       Estimated Output Tokens Per Model:{" "}
                       {info.metadata?.default_estimated_output_tokens_per_model
                         ? JSON.stringify(info.metadata.default_estimated_output_tokens_per_model)
                         : "Default"}
-                    </Text>
+                    </p>
                   </div>
                 </Card>
 
-                <Card>
-                  <Text>Models</Text>
+                <Card className="block p-6">
+                  <p>Models</p>
                   <div className="mt-2 flex flex-wrap gap-2">
                     {computeTeamModelBadges(info.models, info.access_group_models || [], info.access_group_details).map(
                       (badge, index) => (
                         <Tooltip key={`${badge.kind}-${badge.label}-${index}`} title={badge.tooltip}>
                           <span>
-                            <Badge color={TEAM_MODEL_BADGE_COLORS[badge.kind]}>{badge.label}</Badge>
+                            <StatusBadge tone={TEAM_MODEL_BADGE_TONES[badge.kind]} label={badge.label} />
                           </span>
                         </Tooltip>
                       ),
@@ -823,12 +827,12 @@ const TeamInfoView: React.FC<TeamInfoProps> = ({
                   </div>
                 </Card>
 
-                <Card>
-                  <Text className="font-semibold text-gray-900">Virtual Keys</Text>
+                <Card className="block p-6">
+                  <p className="font-semibold text-gray-900">Virtual Keys</p>
                   <div className="mt-2">
-                    <Text>User Keys: {teamData.keys.filter((key) => key.user_id).length}</Text>
-                    <Text>Service Account Keys: {teamData.keys.filter((key) => !key.user_id).length}</Text>
-                    <Text className="text-gray-500">Total: {teamData.keys.length}</Text>
+                    <p>User Keys: {teamData.keys.filter((key) => key.user_id).length}</p>
+                    <p>Service Account Keys: {teamData.keys.filter((key) => !key.user_id).length}</p>
+                    <p className="text-gray-500">Total: {teamData.keys.length}</p>
                   </div>
                 </Card>
 
@@ -838,7 +842,7 @@ const TeamInfoView: React.FC<TeamInfoProps> = ({
                   accessToken={accessToken}
                 />
 
-                <Card>
+                <Card className="block p-6">
                   <GuardrailSettingsView
                     globalGuardrailNames={globalGuardrailNames}
                     teamGuardrails={Array.isArray(info.metadata?.guardrails) ? info.metadata.guardrails : []}
@@ -852,22 +856,22 @@ const TeamInfoView: React.FC<TeamInfoProps> = ({
                   />
                 </Card>
 
-                <Card>
-                  <Text className="font-semibold text-gray-900 mb-3">Policies</Text>
+                <Card className="block p-6">
+                  <p className="font-semibold text-gray-900 mb-3">Policies</p>
                   {info.policies && info.policies.length > 0 ? (
                     <div className="space-y-4">
                       {info.policies.map((policy: string, index: number) => (
                         <div key={index} className="space-y-2">
                           <div className="flex items-center gap-2">
-                            <Badge color="purple">{policy}</Badge>
-                            {loadingPolicies && <Text className="text-xs text-gray-400">Loading guardrails...</Text>}
+                            <Badge variant="secondary">{policy}</Badge>
+                            {loadingPolicies && <p className="text-xs text-gray-400">Loading guardrails...</p>}
                           </div>
                           {!loadingPolicies && policyGuardrails[policy] && policyGuardrails[policy].length > 0 && (
                             <div className="ml-4 pl-3 border-l-2 border-gray-200">
-                              <Text className="text-xs text-gray-500 mb-1">Resolved Guardrails:</Text>
+                              <p className="text-xs text-gray-500 mb-1">Resolved Guardrails:</p>
                               <div className="flex flex-wrap gap-1">
                                 {policyGuardrails[policy].map((guardrail: string, gIndex: number) => (
-                                  <Badge key={gIndex} color="blue" size="xs">
+                                  <Badge key={gIndex} variant="secondary">
                                     {guardrail}
                                   </Badge>
                                 ))}
@@ -878,7 +882,7 @@ const TeamInfoView: React.FC<TeamInfoProps> = ({
                       ))}
                     </div>
                   ) : (
-                    <Text className="text-gray-500">No policies configured</Text>
+                    <p className="text-gray-500">No policies configured</p>
                   )}
                 </Card>
 
@@ -887,7 +891,7 @@ const TeamInfoView: React.FC<TeamInfoProps> = ({
                   disabledCallbacks={[]}
                   variant="card"
                 />
-              </Grid>
+              </div>
             ),
           },
           {
@@ -923,9 +927,9 @@ const TeamInfoView: React.FC<TeamInfoProps> = ({
             key: TEAM_INFO_TAB_KEYS.SETTINGS,
             label: TEAM_INFO_TAB_LABELS[TEAM_INFO_TAB_KEYS.SETTINGS],
             children: (
-              <Card className="overflow-y-auto max-h-[65vh]">
+              <Card className="block p-6 overflow-y-auto max-h-[65vh]">
                 <div className="flex justify-between items-center mb-4">
-                  <Title>Team Settings</Title>
+                  <h3 className="text-lg font-medium">Team Settings</h3>
                   {canEditTeam && !isEditing && (
                     <Button
                       icon={<EditOutlined className="h-4 w-4" />}
@@ -1079,15 +1083,16 @@ const TeamInfoView: React.FC<TeamInfoProps> = ({
                       <Input placeholder="example1@test.com, example2@test.com" />
                     </Form.Item>
 
-                    <Accordion className="mt-4 mb-4">
-                      <AccordionHeader>
+                    <Collapsible className="mt-4 mb-4 overflow-hidden rounded-lg border">
+                      <CollapsibleTrigger className="group/section flex w-full items-center justify-between px-4 py-3 text-left">
                         <b>Team Member Settings</b>
-                      </AccordionHeader>
-                      <AccordionBody>
-                        <Text className="text-xs text-gray-500 mb-4">
+                        <ChevronDown className="size-5 shrink-0 text-gray-500 transition-transform group-data-[panel-open]/section:rotate-180" />
+                      </CollapsibleTrigger>
+                      <CollapsibleContent className="px-4 pb-3">
+                        <p className="text-xs text-gray-500 mb-4">
                           Optional defaults applied when members join this team. All fields can be overridden per
                           member.
-                        </Text>
+                        </p>
                         <Form.Item
                           label={
                             <span>
@@ -1132,7 +1137,7 @@ const TeamInfoView: React.FC<TeamInfoProps> = ({
                           name="team_member_key_duration"
                           tooltip="Set a limit to the duration of a team member's key. Format: 30s (seconds), 30m (minutes), 30h (hours), 30d (days), 1mo (month)"
                         >
-                          <TextInput placeholder="e.g., 30d" />
+                          <UIInput placeholder="e.g., 30d" />
                         </Form.Item>
                         <Form.Item
                           label="Default TPM Limit"
@@ -1148,8 +1153,8 @@ const TeamInfoView: React.FC<TeamInfoProps> = ({
                         >
                           <NumericalInput step={1} style={{ width: "100%" }} placeholder="e.g., 100" />
                         </Form.Item>
-                      </AccordionBody>
-                    </Accordion>
+                      </CollapsibleContent>
+                    </Collapsible>
 
                     <Form.Item label="Reset Budget" name="budget_duration">
                       <BudgetDurationDropdown placeholder="Never resets" />
@@ -1447,11 +1452,12 @@ const TeamInfoView: React.FC<TeamInfoProps> = ({
                       />
                     </Form.Item>
 
-                    <Accordion className="mt-4 mb-4">
-                      <AccordionHeader>
+                    <Collapsible className="mt-4 mb-4 overflow-hidden rounded-lg border">
+                      <CollapsibleTrigger className="group/section flex w-full items-center justify-between px-4 py-3 text-left">
                         <b>Search Tool Settings</b>
-                      </AccordionHeader>
-                      <AccordionBody>
+                        <ChevronDown className="size-5 shrink-0 text-gray-500 transition-transform group-data-[panel-open]/section:rotate-180" />
+                      </CollapsibleTrigger>
+                      <CollapsibleContent className="px-4 pb-3">
                         <Form.Item
                           label="Allowed Search Tools"
                           name="object_permission_search_tools"
@@ -1464,8 +1470,8 @@ const TeamInfoView: React.FC<TeamInfoProps> = ({
                             placeholder="Select search tools (optional, empty = all allowed)"
                           />
                         </Form.Item>
-                      </AccordionBody>
-                    </Accordion>
+                      </CollapsibleContent>
+                    </Collapsible>
 
                     <Form.Item label="Organization" name="organization_id">
                       <Select
@@ -1537,22 +1543,22 @@ const TeamInfoView: React.FC<TeamInfoProps> = ({
                 ) : (
                   <div className="space-y-4">
                     <div>
-                      <Text className="font-medium">Team Name</Text>
+                      <p className="font-medium">Team Name</p>
                       <div>{info.team_alias}</div>
                     </div>
                     <div>
-                      <Text className="font-medium">Team ID</Text>
+                      <p className="font-medium">Team ID</p>
                       <div className="font-mono">{info.team_id}</div>
                     </div>
                     <div>
-                      <Text className="font-medium">Created At</Text>
+                      <p className="font-medium">Created At</p>
                       <div>{new Date(info.created_at).toLocaleString()}</div>
                     </div>
                     <div>
-                      <Text className="font-medium">Models</Text>
+                      <p className="font-medium">Models</p>
                       <div className="flex flex-wrap gap-2 mt-1">
                         {info.models.map((model, index) => (
-                          <Badge key={index} color="red">
+                          <Badge key={index} variant="secondary">
                             {model}
                           </Badge>
                         ))}
@@ -1560,10 +1566,10 @@ const TeamInfoView: React.FC<TeamInfoProps> = ({
                     </div>
                     {info.default_team_member_models && info.default_team_member_models.length > 0 && (
                       <div>
-                        <Text className="font-medium">Default Member Models</Text>
+                        <p className="font-medium">Default Member Models</p>
                         <div className="flex flex-wrap gap-2 mt-1">
                           {info.default_team_member_models.map((model, index) => (
-                            <Badge key={index} color="blue">
+                            <Badge key={index} variant="secondary">
                               {model}
                             </Badge>
                           ))}
@@ -1571,7 +1577,7 @@ const TeamInfoView: React.FC<TeamInfoProps> = ({
                       </div>
                     )}
                     <div>
-                      <Text className="font-medium">Model Aliases</Text>
+                      <p className="font-medium">Model Aliases</p>
                       {(() => {
                         const aliasEntries = Object.entries(info.litellm_model_table?.model_aliases ?? {});
                         if (aliasEntries.length === 0) {
@@ -1591,7 +1597,7 @@ const TeamInfoView: React.FC<TeamInfoProps> = ({
                       })()}
                     </div>
                     <div>
-                      <Text className="font-medium">Rate Limits</Text>
+                      <p className="font-medium">Rate Limits</p>
                       <div>TPM: {info.tpm_limit || "Unlimited"}</div>
                       <div>RPM: {info.rpm_limit || "Unlimited"}</div>
                       {(() => {
@@ -1601,7 +1607,7 @@ const TeamInfoView: React.FC<TeamInfoProps> = ({
                         if (models.length === 0) return null;
                         return (
                           <div className="mt-2">
-                            <Text className="text-gray-500">Per-model limits:</Text>
+                            <p className="text-gray-500">Per-model limits:</p>
                             {models.map((m) => (
                               <div key={m} className="text-xs ml-2">
                                 {m}: TPM {modelTpm[m] ?? "—"}, RPM {modelRpm[m] ?? "—"}
@@ -1619,7 +1625,7 @@ const TeamInfoView: React.FC<TeamInfoProps> = ({
                       </div>
                     </div>
                     <div>
-                      <Text className="font-medium">Team Budget</Text>
+                      <p className="font-medium">Team Budget</p>
                       <div>
                         Max Budget:{" "}
                         {info.max_budget !== null ? `$${formatNumberWithCommas(info.max_budget, 4)}` : "No Limit"}
@@ -1638,12 +1644,12 @@ const TeamInfoView: React.FC<TeamInfoProps> = ({
                         )}
                     </div>
                     <div>
-                      <Text className="font-medium">
+                      <p className="font-medium">
                         Team Member Settings{" "}
                         <Tooltip title="These are limits on individual team members">
                           <InfoCircleOutlined style={{ marginLeft: "4px" }} />
                         </Tooltip>
-                      </Text>
+                      </p>
                       <div>Max Budget: {info.team_member_budget_table?.max_budget || "No Limit"}</div>
                       <div>Budget Duration: {info.team_member_budget_table?.budget_duration || "No Limit"}</div>
                       <div>Key Duration: {info.metadata?.team_member_key_duration || "No Limit"}</div>
@@ -1651,7 +1657,7 @@ const TeamInfoView: React.FC<TeamInfoProps> = ({
                       <div>RPM Limit: {info.team_member_budget_table?.rpm_limit || "No Limit"}</div>
                     </div>
                     <div>
-                      <Text className="font-medium">Router Settings</Text>
+                      <p className="font-medium">Router Settings</p>
                       {info.router_settings &&
                       Object.values(info.router_settings).some(
                         (v) => v !== null && v !== undefined && v !== "" && !(Array.isArray(v) && v.length === 0),
@@ -1659,7 +1665,8 @@ const TeamInfoView: React.FC<TeamInfoProps> = ({
                         <div className="mt-1 space-y-1">
                           {info.router_settings.routing_strategy && (
                             <div>
-                              Routing Strategy: <Badge color="blue">{info.router_settings.routing_strategy}</Badge>
+                              Routing Strategy:{" "}
+                              <Badge variant="secondary">{info.router_settings.routing_strategy}</Badge>
                             </div>
                           )}
                           {info.router_settings.num_retries != null && (
@@ -1687,12 +1694,14 @@ const TeamInfoView: React.FC<TeamInfoProps> = ({
                       )}
                     </div>
                     <div>
-                      <Text className="font-medium">Organization ID</Text>
+                      <p className="font-medium">Organization ID</p>
                       <div>{info.organization_id}</div>
                     </div>
                     <div>
-                      <Text className="font-medium">Status</Text>
-                      <Badge color={info.blocked ? "red" : "green"}>{info.blocked ? "Blocked" : "Active"}</Badge>
+                      <p className="font-medium">Status</p>
+                      <Badge variant={info.blocked ? "destructive" : "secondary"}>
+                        {info.blocked ? "Blocked" : "Active"}
+                      </Badge>
                     </div>
 
                     <ObjectPermissionsView
@@ -1724,7 +1733,7 @@ const TeamInfoView: React.FC<TeamInfoProps> = ({
 
                     {info.metadata?.secret_manager_settings && (
                       <div className="pt-4 border-t border-gray-200">
-                        <Text className="font-medium">Secret Manager Settings</Text>
+                        <p className="font-medium">Secret Manager Settings</p>
                         <pre className="mt-2 bg-gray-50 p-3 rounded-sm text-xs overflow-x-auto">
                           {JSON.stringify(info.metadata.secret_manager_settings, null, 2)}
                         </pre>

ui/litellm-dashboard/src/lib/http/schema.d.ts

@@ -27356,6 +27356,8 @@ export interface components {
             cache_creation_input_token_cost_flex?: number | null;
             /** Cache Creation Input Token Cost Priority */
             cache_creation_input_token_cost_priority?: number | null;
+            /** Cache Creation Input Token Cost Ultrafast */
+            cache_creation_input_token_cost_ultrafast?: number | null;
             /** Cache Read Input Audio Token Cost */
             cache_read_input_audio_token_cost?: number | null;
             /** Cache Read Input Token Cost */
@@ -27376,6 +27378,8 @@ export interface components {
             cache_read_input_token_cost_flex?: number | null;
             /** Cache Read Input Token Cost Priority */
             cache_read_input_token_cost_priority?: number | null;
+            /** Cache Read Input Token Cost Ultrafast */
+            cache_read_input_token_cost_ultrafast?: number | null;
             /** Citation Cost Per Token */
             citation_cost_per_token?: number | null;
             /** Complexity Router Config */
@@ -27440,6 +27444,8 @@ export interface components {
             input_cost_per_token_flex?: number | null;
             /** Input Cost Per Token Priority */
             input_cost_per_token_priority?: number | null;
+            /** Input Cost Per Token Ultrafast */
+            input_cost_per_token_ultrafast?: number | null;
             /** Input Cost Per Video Per Second */
             input_cost_per_video_per_second?: number | null;
             /** Input Cost Per Video Per Second Above 128K Tokens */
@@ -27537,6 +27543,8 @@ export interface components {
             output_cost_per_token_flex?: number | null;
             /** Output Cost Per Token Priority */
             output_cost_per_token_priority?: number | null;
+            /** Output Cost Per Token Ultrafast */
+            output_cost_per_token_ultrafast?: number | null;
             /** Output Cost Per Video Per Second */
             output_cost_per_video_per_second?: number | null;
             /** Output Cost Per Video Token */
@@ -36458,6 +36466,8 @@ export interface components {
             cache_creation_input_token_cost_flex?: number | null;
             /** Cache Creation Input Token Cost Priority */
             cache_creation_input_token_cost_priority?: number | null;
+            /** Cache Creation Input Token Cost Ultrafast */
+            cache_creation_input_token_cost_ultrafast?: number | null;
             /** Cache Read Input Audio Token Cost */
             cache_read_input_audio_token_cost?: number | null;
             /** Cache Read Input Token Cost */
@@ -36478,6 +36488,8 @@ export interface components {
             cache_read_input_token_cost_flex?: number | null;
             /** Cache Read Input Token Cost Priority */
             cache_read_input_token_cost_priority?: number | null;
+            /** Cache Read Input Token Cost Ultrafast */
+            cache_read_input_token_cost_ultrafast?: number | null;
             /** Citation Cost Per Token */
             citation_cost_per_token?: number | null;
             /** Complexity Router Config */
@@ -36542,6 +36554,8 @@ export interface components {
             input_cost_per_token_flex?: number | null;
             /** Input Cost Per Token Priority */
             input_cost_per_token_priority?: number | null;
+            /** Input Cost Per Token Ultrafast */
+            input_cost_per_token_ultrafast?: number | null;
             /** Input Cost Per Video Per Second */
             input_cost_per_video_per_second?: number | null;
             /** Input Cost Per Video Per Second Above 128K Tokens */
@@ -36639,6 +36653,8 @@ export interface components {
             output_cost_per_token_flex?: number | null;
             /** Output Cost Per Token Priority */
             output_cost_per_token_priority?: number | null;
+            /** Output Cost Per Token Ultrafast */
+            output_cost_per_token_ultrafast?: number | null;
             /** Output Cost Per Video Per Second */
             output_cost_per_video_per_second?: number | null;
             /** Output Cost Per Video Token */