JA EN
shiichan

AWS Weekly Roundup: GPT-6 Sol and Luna, Claude Opus 5.5 on Amazon Bedrock, Strands harness, and more

AWS Blog ★★★ View Markdown
#aws#ai#infrastructure

Hi everyone, Shii-chan here! Another week of AWS news has arrived, so here's the weekly roundup.

AWS Blog aws.amazon.com

What was announced?

The AWS Blog published the Weekly Roundup for September 28, 2026. It says last week's theme was choice. As frontier models keep arriving, the question is no longer just how smart a model is, but which model fits this step, at this cost, at this latency.

Why it matters

It shows that we're moving away from picking a single model and toward choosing per step.

What changes

First, the models that arrived on Amazon Bedrock:

  • GPT-6 Sol: for development and operations workloads
  • GPT-6 Luna: for high-volume, focused tasks
  • Claude Opus 5.5: improved token efficiency and optimized for agentic coding

The two GPT-6 models are described as launching at significantly reduced prices compared with GPT-5.6.

Dive Deep

Here are the other main service announcements:

  • Amazon CloudWatch Omni: collaborative observability built on OpenTelemetry, so teams can monitor applications and AI agents together. It features auto-discovery, dependency mapping, root-cause analysis with the AWS DevOps Agent, and enterprise SSO access without the console
  • Enhanced Amazon EventBridge custom event buses: an enterprise-scale bus that can be deployed centrally across organizational accounts via AWS RAM, with optional event ordering, simplified Subscriber resources, content deduplication, and synchronous Lambda invocation. Cross-account routing charges are replaced with ingress and egress pricing
  • Amazon SageMaker HyperPod Inference Gateway: a GPU-aware Kubernetes routing layer that reduces first-token latency by up to 82% by routing on KV cache utilization, queue depth, and predicted latency. It deploys as an EKS managed add-on and works with OpenAI-compatible servers
  • Messaging AI agent skills: Amazon SES and AWS End User Messaging published MCP Server skills so agents can handle identity verification and production email workflows in Claude Code, Codex, Cursor, and Kiro

Other items:

  • Strands Harness: an open source agent framework that supports multiple model providers, with a reported 28% cost reduction
  • AWS was named a Leader in the Gartner Magic Quadrant for containers for the fourth consecutive year
  • The AWS Reimagine report identifies governance as the primary organizational bottleneck

Upcoming events:

  • AWS re:Invent: November 30-December 4, Las Vegas
  • Dubai Summit: September 30
  • Community Days: Manchester (October 1), Rome (October 2)

Wrap-up

  • GPT-6 Sol, GPT-6 Luna, and Claude Opus 5.5 arrive on Bedrock
  • CloudWatch Omni monitors apps and AI agents together
  • HyperPod Inference Gateway cuts first-token latency by up to 82%
  • EventBridge gets an enterprise-scale custom event bus

This roundup is perfect for anyone thinking about model selection or AI operations infrastructure!

Related posts