AWS AI Updates: July 2, 2026

1. Bedrock AgentCore Raises Default Runtime Quotas

AWS. Amazon Bedrock AgentCore increased its default runtime quotas so teams can run more AI agents concurrently and handle higher-throughput workloads without requesting limit increases. The new defaults allow up to 5,000 active concurrent sessions in US East (N. Virginia) and US West (Oregon) and 2,500 in other supported regions, plus 200 agent interactions per second and 25 new sessions created per second across all regions. AWS positions the change as removing manual quota tuning for production agent deployments that need substantial concurrent capacity. Source

2. AppConfig Adds Managed Experimentation for AI Model and Prompt Testing

AWS. AWS AppConfig launched managed experimentation tools for running A/B tests and multivariate experiments across the application stack, explicitly including AI model selections and prompt experiments alongside UI changes and recommendation algorithms. The tooling includes intelligent validation that checks experiment setups against Amazon’s best practices to ensure sufficient statistical power, and it works across EC2, Lambda, ECS, EKS, and on-premises servers via the AppConfig Agent. It is available in all AWS Regions, including GovCloud (US), with pay-as-you-go pricing based on experimentation hours. Source

3. GPT-5.4 and Nemotron 3 Super 120B Arrive on Kiro in GovCloud

AWS. OpenAI GPT-5.4 and NVIDIA Nemotron 3 Super 120B are now selectable in the Kiro IDE and CLI in the AWS GovCloud (US-West) Region, served through Amazon Bedrock’s inference engine. GPT-5.4 targets complex reasoning, coding, document analysis, and multi-step agentic workflows with a 272K context window at a 1.2x credit multiplier, while Nemotron 3 Super is an open-weight hybrid mixture-of-experts model that activates only 12B of its 120B parameters, offering a 256K context window and a lower 0.25x credit multiplier for cost efficiency. Developers access the models by updating their IDE or CLI to the latest version and selecting them from the model picker. Source

4. Claude Opus 4.8 Now Available in AWS GovCloud (US)

AWS. Anthropic’s Claude Opus 4.8, described by AWS as Anthropic’s most capable generally available model to date, is now accessible in AWS GovCloud (US) through Amazon Bedrock. The model targets coding across large codebases with planning before edits and long-session context retention, autonomous task execution that recovers from its own errors, and knowledge work that synthesizes across long documents and self-checks output. GovCloud access comes with AWS-managed features including Guardrails, Knowledge Bases, and regional data residency protections. Source