AWS AI Updates: August 25, 2026

1. OpenAI GPT-5.6 Terra and Luna Land on Bedrock in GovCloud

AWS. OpenAI’s GPT-5.6 Terra and Luna models are now generally available on Amazon Bedrock in AWS GovCloud (US-West) and GovCloud (US-East), with support for 1 million token context windows. Terra delivers GPT-5.5-level capability at lower cost while Luna targets fast, low-price inference, and both support prompt caching with explicit breakpoints billed at a 90 percent discount for repeated context in agentic workflows. Access is available through the Bedrock Console and the Responses API on the bedrock-mantle endpoint for government and regulated workloads. Source

2. SageMaker HyperPod Expands Ray Framework Support

AWS. Amazon SageMaker HyperPod added four Ray enhancements: Grafana dashboards backed by Amazon Managed Service for Prometheus plus one-click access to the Ray Dashboard, automatic node recovery and hung job detection with tiered checkpointing that restores state from cluster memory, and a tiered key-value cache that reuses cached prefixes to cut time to first token when serving with Ray Serve. It also lets data scientists attach JupyterLab, Code Editor, or a local IDE to a running Ray cluster for interactive iteration. The updates target resilient distributed training and accelerated inference on GPU clusters. Source

3. Amazon Connect Adds AI-Powered Information Extraction

AWS. Amazon Connect now uses AI to automatically capture data from voice and chat interactions, extracting both verbatim values such as account numbers and reservation IDs and derived insights such as reason for contact, resolution provided, and next steps promised. The system operates on raw contact content before redaction, so teams can extract needed data points while still redacting sensitive values. Extracted data is available to agents during after-contact work, to supervisors via contact search, and to developers through APIs, Kinesis Data Streams, and S3, enabling automated downstream actions like task creation and email notifications. Source

4. SageMaker MLflow Adds Customer-Managed KMS Key Encryption

AWS. Amazon SageMaker MLflow now supports encryption with customer-managed keys (CMK) through AWS KMS, letting organizations control their own encryption rather than relying on AWS-managed keys. The implementation requires a symmetric KMS key created in the same account and region as the MLflow application, and integrates with AWS CloudTrail for audit trails of data access. The capability is generally available in all regions where the MLflow App operates. Source