Catch unexpected spending from third-party models on Amazon Bedrock

AWS Cost Anomaly Detection now monitors third-party foundation models running on Amazon Bedrock, including provider-hosted models such as Anthropic Claude.

Alerts include root-cause breakdowns by account, Region, service, and usage type. That gives teams more detail when investigating unexpected generative artificial intelligence spending.

Lower the price of high-volume GPT-5.6 Sol workloads

Amazon Bedrock reduced pricing for OpenAI GPT-5.6 Sol to $4 per million input tokens and $20 per million output tokens.

The reductions are 20% for input tokens and 33.3% for output tokens. Promotional pricing is available at least through November 21, 2026.

Make SageMaker inference deployments easier to compare by cost

Generative AI Inference Recommendations are now available in SageMaker AI Studio.

The guided benchmarking workflow ranks inference configurations by latency, throughput, and cost. Teams can compare instance types, serving containers, and optimization strategies using real workloads.

Save 30% on AWS Glue 6.0 while gaining Iceberg v3 support

AWS Glue 6.0 is generally available with a 30% price reduction, along with Apache Spark 4.1 and performance improvements for batch, streaming, extract, transform, and load, and artificial intelligence workloads.

The release also adds Apache Iceberg v3 capabilities and more efficient processing. Together, the lower service price and processing improvements can reduce service costs and operational overhead.

Improve cost allocation for centralized CloudWatch logs

Amazon CloudWatch Centralization now propagates source log group tags to destination log groups.

The copied tags can include ownership and cost-center information. This supports more accurate cost allocation and reporting when logs are centralized across accounts and Regions.

Identify expensive Aurora DSQL statements with CloudWatch Database Insights

Amazon Aurora DSQL now provides per-statement, cluster-level performance metrics through CloudWatch Database Insights at no additional charge.

Teams can identify resource-intensive queries and use the information to tune workloads. The statement-level view also supports database cost optimization by showing which queries are consuming resources.

Manage more reservation types through EC2 Capacity Reservation Resource Groups

AWS Capacity Reservation Resource Groups now support EC2 Capacity Blocks and interruptible reservations.

Resource groups can combine On-Demand Capacity Reservations, interruptible reservations, and Capacity Blocks for machine learning. EC2 Fleet and Auto Scaling can prioritize reservation types and fall back to On-Demand capacity.

Run AWS Batch workloads on ECS Managed Instances

AWS Batch now supports Amazon ECS Managed Instances for GPU-accelerated and compute-intensive jobs.

AWS handles Amazon Machine Image updates, patching, and instance lifecycle operations. Batch environments can use On-Demand, Spot, or reserved capacity, providing additional ways to balance workload efficiency and compute cost.