Amazon announced three distinct updates this week: GPT‑6.1 Terra is now generally available on Amazon Bedrock with reduced token costs; Step Functions workflows can be invoked directly by AI agents via the Bedrock AgentCore Gateway; and a set of post‑quantum and identity‑propagation features landed across KMS, Aurora PostgreSQL, IAM Identity Center, ECS, and Lambda. Each change alters the engineering surface for AI, platform, DevOps, and security teams.
GPT‑6.1 Terra on Bedrock
OpenAI’s Terra variant of the GPT‑6.1 family is now accessible through the Bedrock APIs. The model supports up to 1 million input tokens and accepts both text and image payloads. OpenAI reports that Terra’s input and output token pricing is at least 50 % lower than the comparable GPT‑6.1 Sol and GPT‑5.6 Terra offerings, and that it completes tasks more than three times faster than those models. While the performance numbers are vendor‑provided, the pricing shift means cost‑sensitive workloads—such as batch summarisation or multi‑modal inference—can be re‑evaluated for Bedrock deployment.
Step Functions Exposed as AI Agent Tools
Step Functions, which already orchestrates over 220 AWS services, is now exposed to generative agents through the Bedrock AgentCore Gateway using the Model Context Protocol (MCP). The agent decides when to invoke a workflow and supplies the required input; Step Functions then executes each defined state in the original order. Existing Standard and Express workflows are eligible without code changes, and agents can also query prior runs. This separation of decision‑making (agent) and deterministic execution (workflow) enables teams to add AI‑driven automation to established processes—such as refund handling or incident response—without rewriting orchestration logic.
Post‑Quantum and Cross‑Region Identity Enhancements
AWS KMS now offers ML‑KEM for post‑quantum key encapsulation, allowing a shared secret to be derived via Encapsulate and recovered with Decapsulate. The feature supports all three NIST FIPS 203 parameter sets and is available in every KMS region except the UAE, Bahrain, and China. In parallel, Aurora PostgreSQL (versions 18+) added post‑quantum TLS key exchange, extending the same cryptographic hardening to database connections. Additionally, IAM Identity Center introduced cross‑Region trusted identity propagation, letting a user’s identity travel with requests to services in other regions, eliminating the need for broad IAM roles or single‑region consolidation.
Compute and Runtime Optimisations
Amazon ECS now supports GPU sharing via NVIDIA Multi‑Process Service (MPS). Containers can declare required GPU memory and optional compute caps, allowing multiple tasks to co‑locate on a single GPU. This reduces the need for over‑provisioned GPU instances for low‑intensity inference workloads. The feature works on both ECS Managed Instances and EC2‑backed clusters, with health monitoring that replaces unhealthy instances. AWS Lambda added support for Seekable OCI (SOCI) lazy loading on Managed Instances, enabling the runtime to fetch only the files needed at each startup phase, which can shorten cold‑start latency and lower data transfer costs.
Related CloudNinjas coverage: AWS.
What This Means For Practitioners
- Re‑evaluate model selection for cost‑sensitive AI pipelines; Terra’s lower token pricing may shift workloads from older models.
- Leverage the AgentCore Gateway to wrap existing Step Functions workflows with AI decision logic, preserving deterministic execution while adding flexibility.
- Plan for post‑quantum cryptography adoption by migrating KMS keys to ML‑KEM and enabling Aurora’s TLS option where compliance or long‑term data confidentiality is a concern.
- Use IAM Identity Center’s cross‑Region propagation to simplify multi‑region architectures without expanding IAM role scopes.
- Adopt ECS GPU sharing for inference services that do not fully utilise a GPU, and enable SOCI in Lambda to reduce cold‑start overhead for container‑based functions.
Monitoring the adoption metrics of these features and testing them in staging environments will help teams decide when to incorporate them into production pipelines.