The latest AI news and updates of AWS — Cloud platform offering AI services including Bedrock, SageMaker, and managed inference. Covering AWS's latest product updates, research, and launches from the past 90 days.
Amazon Web ServicesOct 8AWS has made Anthropic’s Claude Haiku 5.5 available on Amazon Bedrock and Claude Platform on AWS. This model, the fastest in the Claude 5.5 family, supports subagents and high-volume, cost-sensitive tasks. It features effort controls and costs approximately 75% less than Claude Haiku 4.5 for most workloads, while handling coding, tool use, and computer use.
Amazon Web ServicesSep 30AWS launched general availability for OpenAI's GPT-6.1 Sol on Amazon Bedrock. The model delivers near-Astra reasoning performance for agentic coding, computer use, and professional workflows. It achieves these results at one-fifth the cost per task of previous frontier models, while providing enterprise-grade security controls like zero-operator access and VPC endpoints for production deployments.
Amazon Web ServicesSep 25AWS released the Deception Benchmark, a dataset of 14,822 code samples designed to test whether AI models can distinguish real vulnerabilities from safe code that appears suspicious. Evaluations of 12 models from five providers show that while models catch nearly all real vulnerabilities, they also flag safe code as dangerous up to 99% of the time.
Amazon Web ServicesSep 23AWS has made OpenAI's GPT-6 Sol and GPT-6 Luna generally available on Amazon Bedrock. GPT-6 Sol handles complex recurring tasks like coding and data analysis, while GPT-6 Luna optimizes high-volume, focused workloads. Both models support explicit prompt caching to reduce latency, and Bedrock provides enterprise-grade security controls including zero-operator access and data encryption for all model invocations.
Amazon Web ServicesSep 23AWS introduced Amazon CloudWatch Omni, an AI-powered observability experience that unifies application and AI agent telemetry in one place. The platform operates outside the AWS Console, offering auto-discovered topology, natural-language querying, and IDE extensions for VS Code, Cursor, and Kiro. It supports OpenTelemetry and correlates agent reasoning with application and infrastructure health to identify root causes.
Amazon Web ServicesSep 22AWS has made Anthropic's Claude Opus 5.5 available on Amazon Bedrock and Claude Platform on AWS. This model, Anthropic's most capable Opus version to date, offers a lower cost per token and is designed for long-running coding and knowledge work. It features adaptive thinking and new safety classifiers for biology, cybersecurity, and AI development.
Amazon Web ServicesSep 22AWS now hosts Moonshot AI's Kimi K3, an open-weight model featuring 2.8 trillion parameters, native vision, and a 1-million-token context window. The integration supports explicit prompt caching to reduce latency and input costs. Kimi K3 runs within the AWS data boundary, maintaining zero data retention and zero operator access for enterprise-grade security during inference.
Amazon Web ServicesSep 15AWS has launched seven new agentic AI solutions from its Partner Agent Factory program in the AWS Marketplace. These agents, built by AWS partners using Amazon Bedrock and AgentCore, cover domains including clinical documentation, video intelligence, and financial services. The Marketplace listing provides simplified procurement and consolidated billing that applies toward existing AWS spending commitments.
Amazon Web ServicesSep 9AWS launched general availability for OpenAI's GPT-6 Astra on Amazon Bedrock. The model supports complex reasoning and computer-use workflows across enterprise applications. Amazon Bedrock provides enterprise security controls, including zero-operator access and data encryption, while new enterprise plugins extend browser-use capabilities across business applications.
Amazon Web ServicesSep 2AWS now offers Anthropic's Claude Fable 5.1 on Amazon Bedrock and Claude Platform on AWS. The model supports complex coding, research, and enterprise reasoning tasks. It carries Covered Model status, requiring 30-day data retention with human review, while Enterprise Frontier Safeguards provide zero data retention for eligible deployments through December 31, 2026.
Amazon Web ServicesAug 28AWS updated Web Search on Amazon Bedrock AgentCore to version 1.2.0, adding runtime domain and published-date filtering. The update enforces per-request source and time-window restrictions server-side. Additionally, the tool is now available in the Dublin (eu-west-1) and Tokyo (ap-northeast-1) regions, supporting zero-egress architectures for regulated workloads.
Amazon Web ServicesAug 25AWS now offers managed Ray capabilities on Amazon SageMaker HyperPod, providing a complete environment for distributed AI training and serving. The integration includes built-in observability, automatic node recovery, hung job detection, and tiered checkpointing for resilient training. It supports standard Ray APIs and KubeRay, allowing existing scripts to run on EKS-orchestrated clusters without manual infrastructure management.
Amazon Web ServicesAug 22AWS now supports OpenAI GPT-5.4 on Amazon Bedrock in AWS GovCloud (US). The model carries FedRAMP High and DoD IL4/5 authorization, allowing federal agencies to integrate frontier AI within their compliance boundaries. The service uses the OpenAI-compatible Responses API through the regional Mantle endpoint, with configurable reasoning effort and structured output support for incident review workflows.
Amazon Web ServicesAug 22AWS launched AgentCore payments, enabling AI agents to autonomously pay for APIs, MCPs, and content. The service integrates with Coinbase and Stripe Privy wallets, enforces infrastructure-level spending limits per session, and provides end-to-end observability. It supports the x402 and Machine Payment Protocol, allowing agents to handle pay-per-use and dynamic pricing models without human intervention.
Amazon Web ServicesAug 8AWS launched new guardrails for Amazon Bedrock AgentCore, introducing temporal policies and gateway-level rate limiting. Temporal policies, powered by the new open-source Dogwood language, evaluate sequences of agent actions to enforce constraints like budget caps or required step orders. Gateway rate limiting caps token consumption and request volume per user across all agents without requiring changes to agent code.
Frequently asked questions