AWS Data Exports now provides standardized Amazon Bedrock product metadata
FinOps teams can now slice Bedrock costs by model, token type, and inference mode natively in CUR 2.0—no custom parsing required.
View original announcement →Visual Summary
graph TD
A{{Standardized Bedrock Metadata}}:::announced
B(Amazon Bedrock):::compute
C(AWS Data Exports):::compute
D(Amazon S3):::storage
E(Amazon Athena):::compute
F([Product Map Attributes]):::feature
G([Cost Attribution]):::feature
H((FinOps Teams)):::external
B ==>|"generates usage"| C
C ==>|"standardized fields"| A
A -->|"delivers exports"| D
D -->|"queries"| E
A -->|"exposes"| F
F -->|"model, provider, type"| G
H -->|"analyzes costs"| E
H -.->|"filters & allocates"| G
classDef announced fill:#ff9900,stroke:#ec7211,color:#fff,font-weight:bold
classDef compute fill:#e3f2fd,stroke:#1565c0,color:#1565c0
classDef storage fill:#e8f5e9,stroke:#2e7d32,color:#2e7d32
classDef feature fill:#fff3e0,stroke:#e65100,color:#e65100
classDef external fill:#f5f5f5,stroke:#616161,color:#616161
What's New
AWS has introduced standardized product metadata for Amazon Bedrock within AWS Data Exports (CUR 2.0), giving FinOps teams and cloud administrators consistent, structured attributes to analyze and allocate Bedrock costs. The new fields include model provider, model name, pricing unit, inference type (e.g., input tokens, output tokens), and feature (e.g., On-Demand or Batch inference mode), along with a unified "Amazon Bedrock" product family name that consolidates all Bedrock line items. These attributes are available by default at no additional cost to all Amazon Bedrock customers using AWS Data Exports.
How It Works
- CUR 2.0 product map column: The attributes
model_provider,model_name,inference_type, andfeatureare surfaced as key-value pairs inside theproductmap column in CUR 2.0, queryable individually using the dot operator (e.g.,product.model_name). - Dedicated pricing unit column: The
product_pricing_unitfield is exposed as a standalone column, making it directly accessible without parsing nested map structures. - Unified product family: All Amazon Bedrock line items now share the
product_product_familyvalue of "Amazon Bedrock," enabling a single filter to capture all Bedrock-related costs across models and features. - Inference type granularity: The
inference_typeattribute distinguishes between input tokens and output tokens, allowing cost attribution at the token-direction level for each model invocation. - Feature/serving mode attribute: The
featureattribute identifies the inference serving mode (On-Demand vs. Batch), enabling cost comparisons between workload execution strategies. - Delivery pipeline: Exports are delivered to Amazon S3 and can be queried with Amazon Athena or loaded into a data warehouse, fitting into existing FinOps and analytics pipelines without additional tooling.
Why It's Important
- Eliminates custom parsing logic: Previously, teams had to write bespoke ETL code to extract and normalize Bedrock cost dimensions from inconsistent or unstructured CUR metadata; standardized fields remove that burden entirely.
- Enables accurate cost attribution: With model-level and inference-type granularity, organizations can allocate AI spending to specific teams, applications, or business units using standard tagging and filtering workflows.
- Supports FinOps maturity for AI workloads: As Bedrock adoption scales across enterprises, having structured cost data is a prerequisite for unit economics analysis, budgeting, and forecasting of generative AI spend.
- Facilitates model cost comparison: By exposing model provider and model name as discrete fields, teams can directly compare the cost efficiency of different foundation models (e.g., Anthropic Claude vs. Meta Llama) for the same workload.
- Enables On-Demand vs. Batch cost analysis: The
featureattribute allows teams to quantify savings from shifting workloads to Batch inference, supporting data-driven decisions on inference strategy. - No additional cost or configuration: The fields are available by default, lowering the barrier to adoption and ensuring all Bedrock customers benefit immediately.
How It's Different
- Structured vs. unstructured metadata: Prior to this update, Bedrock cost metadata in CUR was inconsistent and required custom parsing; the new standardized schema provides a reliable, documented contract for downstream consumers.
- Unified product family vs. fragmented line items: Previously, Bedrock costs could appear under multiple product family values, complicating aggregation; the single "Amazon Bedrock" family name simplifies total-spend queries to a single filter.
- Native CUR 2.0 integration vs. third-party workarounds: The attributes are built directly into the AWS Data Exports schema, eliminating the need for third-party cost management tools or manual enrichment pipelines to achieve the same visibility.
- Token-direction granularity: Unlike generic compute cost reports, these fields distinguish input from output token costs, a level of granularity unique to LLM pricing models and not previously available natively in CUR.
- Serving-mode awareness: The
featureattribute surfaces inference mode (On-Demand vs. Batch) as a first-class cost dimension, which is not a concept present in traditional AWS service cost reports.
When to Prefer It
- Multi-model Bedrock deployments: When an organization uses multiple foundation models from different providers, standardized metadata enables side-by-side cost comparisons without custom normalization.
- Chargeback and showback programs: When FinOps teams need to allocate AI costs to internal business units or product teams, structured model-level attributes provide the granularity required for accurate chargebacks.
- Batch vs. On-Demand optimization analysis: When evaluating whether to migrate latency-tolerant workloads from On-Demand to Batch inference, the
featureattribute enables direct cost comparison in existing BI dashboards. - Automated cost anomaly detection: When building automated alerting pipelines on top of CUR data, consistent field names and values reduce false positives caused by metadata inconsistencies.
- Executive AI spend reporting: When preparing board-level or leadership reports on generative AI investment, a unified "Amazon Bedrock" product family simplifies roll-up queries and avoids undercounting.
- Data warehouse or Athena-based cost analytics: When teams already query CUR 2.0 data via Athena or a data warehouse, the new columns integrate seamlessly into existing schemas and dashboards without pipeline changes.
Availability
- General Availability: The standardized Bedrock product metadata is generally available as of July 20, 2026.
- Pricing: Available at no additional cost to all Amazon Bedrock customers using AWS Data Exports (CUR 2.0).
- Enablement: Fields are available by default; no opt-in configuration or export customization is required to access them.
- CUR version requirement: The structured attributes are specific to CUR 2.0 (AWS Data Exports); legacy CUR users would need to migrate to CUR 2.0 to access the standardized fields.
- Regional availability: No region-specific restrictions were noted in the announcement; availability is expected to follow standard AWS Data Exports and Amazon Bedrock regional coverage.
- Access path: Data is delivered to Amazon S3 and queryable via Amazon Athena or loadable into a data warehouse; no direct console visualization is included in this announcement.
