OpenAI GPT-6.1 Terra arrives on Bedrock; Step Functions become agent tools

This week’s most significant developments in artificial intelligence focus less on raw model capability and more on accessibility and cost-efficiency for enterprise workloads. OpenAI has made its GPT-6.1 Terra model generally available on Amazon Bedrock at a price point significantly lower than its predecessor, while AWS Step Functions workflows are now accessible as tools for AI agents through the Model Context Protocol.

New Agentic Capabilities

The integration of existing infrastructure with new agent technologies allows organizations to leverage trusted workflows without major rewrites. AWS Step Functions, which orchestrates over 220 AWS services, can now be handed directly to an agent as a tool via Amazon Bedrock AgentCore Gateway. This setup ensures that complex multi-step processes execute exactly as defined by engineers rather than being improvised by the model.

In parallel, MoEngage demonstrated how they exposed roughly 400 internal API operations to agents using similar technology. By allowing agents to search for tools based on task descriptions instead of providing a full catalog, companies can manage massive tool listings efficiently without forcing models to process thousands of tokens per connection.

OpenAI GPT-6.1 Terra Pricing and Performance

OpenAI’s newest model, the GPT-6.1 Terra line, is positioned as the standard for everyday workloads within its family. According to OpenAI's own "Agents' Last Exam" evaluation, this model scored 15 percentage points higher than GPT-5.6 Terra and performed on par with GPT-6 Astra. It also completed tasks more than three times faster than both models it was compared against.

The primary driver for adoption appears to be cost efficiency. Input and output tokens are priced at least 50% lower than the adjacent GPT-6.1 Sol model, making it a viable option for high-volume applications. The model supports up to one million input tokens with text and image data and is accessible through Bedrock APIs or ChatGPT Work and Codex.

Post-Quantum Security Advances

Security updates this week address the looming threat of quantum computers breaking current encryption standards, specifically through "harvest now, decrypt later" attacks where adversaries record key exchanges today to decode them in the future. AWS Key Management Service (KMS) now supports ML-KEM for post-quantum key encapsulation using private keys validated under FIPS 140-3 Level 3 standards.

These security enhancements extend across multiple services:

• AWS KMS: Supports all three NIST FIPS 203 parameter sets via Encapsulate and Decapsulate operations, available in all regions except Middle East (UAE and Bahrain) and China.
• Amazon Aurora PostgreSQL: Added post-quantum TLS key exchange support on versions 18 and higher.

Infrastructure Efficiency and Scaling

Beyond security, this week saw several updates focused on reducing latency and optimizing resource utilization. Amazon ECS now supports GPU sharing with NVIDIA Multi-Process Service (MPS), enabling multiple tasks to run concurrently on a single NVIDIA GPU. This feature helps organizations utilize underused hardware for small-model inference and multi-model serving without requiring code changes.

Similarly, AWS Lambda introduced Seekable OCI (SOCI) lazy loading for Managed Instances. Instead of pulling an entire container image at startup, the system fetches only the files needed at each stage, reducing startup times by up to 70%. This technique has also been applied to Amazon SageMaker AI training and processing jobs, where lazy loading reduced median startup times from approximately three minutes and 42 seconds down to a range of three to 11 seconds.

Identity Management and Networking

Amazon IAM Identity Center now supports cross-Region trusted identity propagation, removing the architectural constraint that previously forced services accessing a specific user's identity to reside in the same region as that user. This allows identities to travel with requests automatically, enabling granular access controls based on user and group membership regardless of service location.

On the networking front, Amazon EKS Auto Mode now supports Kubernetes Gateway API. This allows traffic routing declarations using HTTPRoute resources without installing a separate controller. The feature provides automatic TLS certificate discovery through AWS Certificate Manager and supports various routing types including cross-namespace routing and weighted traffic splitting while maintaining backward compatibility with existing Ingress workloads.

Real-World Use Cases

Nubank, serving over 140 million customers across Brazil, Mexico, and Colombia, utilized the AWS DevOps Agent to automate disaster recovery testing for critical business flows. By delegating evidence collection and report writing to the agent, Nubank reduced a test cycle from about 4.3 hours to roughly 40 minutes while quadrupling flow coverage in 2026 to more than 160 scenarios.

In storage optimization, an engagement described how moving petabyte-scale logs from Amazon S3 Standard to Glacier Instant Retrieval cut monthly costs by approximately 70%. The data, totaling 5.8 PB with access ratios below 0.1% past 90 days, was transitioned using evidence-based lifecycle rules generated via S3 Storage Lens groups and Storage Class Analysis to ensure safe archival of cold data.