AWS Weekly Roundup: Connect Talent GA, Corretto 27, Kimi K3 on Bedrock, and T8i Instances — Sept 21, 2026
Cloud Platform EngineeringFive GA announcements and two notable additions in this week’s AWS roundup. The Java team, the AI/ML platform team, and the compute team all shipped — with Corretto 27’s post-quantum TLS and Kimi K3’s explicit prompt caching standing out for CPE teams tracking infrastructure decisions.
Amazon Connect Talent — GA
AI agents now conduct structured voice interviews, administer competency assessments, and score candidates through Amazon Connect. The GA release includes adaptive questioning, a mobile-first candidate portal, and anonymized candidate data during evaluation. For engineering teams running high-volume hiring pipelines, Connect Talent is a managed alternative to building interview-scheduling and screening automation on top of existing telephony infrastructure.
Amazon Corretto 27 — GA
Corretto 27 ships on Linux, Windows, and macOS with support through April 2027. Three changes worth noting for Java platform teams:
G1 becomes the default garbage collector. Teams not explicitly setting GC will inherit G1 behavior. For applications currently running on Parallel GC by default, this is a configuration change to audit before upgrading.
Post-quantum hybrid key exchange for TLS 1.3. This is the compliance-ahead-of-mandate move — hybrid algorithms run both classical and post-quantum key exchange simultaneously, protecting in-flight traffic against future quantum decryption of captured sessions. Migration-ready Java infrastructure now.
Compact object headers. Reduced memory footprint per object, which translates to heap efficiency at scale for applications with large object graphs.
Kimi K3 on Amazon Bedrock
Moonshot AI’s Kimi K3 is now accessible through Bedrock — the first open-weight model on the platform to support explicit prompt caching. Key numbers: 1M token context window, native vision capabilities, approximately 2.5× improved scaling efficiency over K2. For CPE teams managing multi-tenant inference infrastructure where prompt caching directly affects cost-per-request, explicit caching control in a 1M-token model is a new architectural option that Bedrock’s existing lineup doesn’t offer.
EC2 T8i — Burstable Instances
T8i instances use sixth-generation Intel Xeon processors, delivering 30% better price-performance over T3. These are burstable workloads — the sweet spot remains development environments, low-traffic applications, and workloads with predictable burst patterns. The T8i refresh is a cost-optimization opportunity for teams still on T3 without a business reason to be there.
Elastic Beanstalk Cluster Mode
Multiple applications now run through one operational baseline on shared Amazon EKS infrastructure. Beanstalk Cluster Mode supports Java, .NET, Python, Node.js, PHP, Ruby, and Go with automatic containerization via Cloud Native Buildpacks. For platform teams that have inherited Beanstalk-managed applications but are standardizing on EKS, this is a migration path that reduces per-application operational overhead without a full replatform.
AgentCore Runtime
Amazon Bedrock’s AgentCore Runtime received an update delivering consistent cold start times and session-based memory reclamation. Billing now aligns with actual usage rather than peak consumption. For teams running Bedrock agents at variable load, this reduces the cost overhang from sessions that complete early.
The So What
The Corretto 27 + post-quantum TLS combination is the change to act on most deliberately — it’s a proactive infrastructure decision that becomes easier to make now than when post-quantum compliance becomes a mandate. For teams evaluating 1M-context models on Bedrock, Kimi K3’s explicit prompt caching adds a cost control lever that has been missing from large-context inference options.
Source: AWS News Blog
Content created with AI assistance and reviewed for accuracy.
Join the conversation
Stack Insiders is our free community for readers who want to go deeper — share resources, ask questions, and connect with others across every vertical we cover.
Join Stack Insiders →