AWS AI Updates: June 2, 2026

1. GPT-5.5, GPT-5.4, and Codex from OpenAI Now Generally Available on Amazon Bedrock

AWS. Amazon Bedrock has added GPT-5.5, GPT-5.4, and OpenAI’s Codex as generally available models, bringing OpenAI’s latest offerings under AWS governance and security controls. GPT-5.5 runs on Bedrock’s next-generation inference engine and is positioned for agentic coding, data analysis, and multi-step autonomous tasks, while Codex is accessible via the Codex App, CLI, and IDE plugins for VS Code, JetBrains, and Xcode. AWS customers can apply existing commitments toward Codex usage, with pricing aligned to OpenAI’s standard rates. Source

2. Amazon Bedrock Adds CloudWatch Metrics for OpenAI- and Anthropic-Compatible APIs

AWS. Amazon Bedrock now publishes CloudWatch metrics under the AWS/BedrockMantle namespace for its bedrock-mantle endpoint, which supports the OpenAI Responses API, OpenAI Chat Completions API, and the Anthropic Messages API. Customers can monitor inference request counts, input/output token totals, and client error counts at account, project, model, and combined project-model granularity levels, enabling cost attribution and capacity planning. The feature allows teams migrating existing OpenAI or Anthropic applications to Bedrock with minimal code changes to gain full operational visibility through standard AWS tooling. Source

3. SageMaker HyperPod Gains AI Coding Assistant Troubleshooting Skills

AWS. Amazon SageMaker HyperPod now offers open-source troubleshooting skills that integrate with AI coding assistants — including Claude Code, Cursor, and Kiro — to diagnose and resolve large-scale ML cluster issues through natural language. The skills encode AWS best practices into structured diagnostic workflows covering GPU fault detection, NCCL communication failures, performance bottleneck identification, version drift detection, and overall cluster health validation. Available for both Slurm and Amazon EKS-orchestrated HyperPod clusters, the skills are published to the AWSLabs GitHub repository via the SageMaker AI plugin. Source

4. SageMaker HyperPod Supports EFA-Only Network Interfaces for Large-Scale Training

AWS. Amazon SageMaker HyperPod now supports configuring network interfaces exclusively for Elastic Fabric Adapter (EFA) traffic without requiring an associated ENA device, removing a scaling bottleneck for distributed AI/ML training clusters. Previously, each EFA interface consumed a VPC subnet IP address via a paired ENA device, constraining the maximum cluster size within a single subnet; EFA-only interfaces eliminate that IP exhaustion problem entirely. Users enable the feature by specifying efa-only in the ClusterNetworkInterface configuration when creating or updating HyperPod clusters through the AWS API. Source