What's New with AWS — June 2026: The Month in Review
A themed roundup of every headline launch AWS shipped in June 2026 — the AWS Summit New York City keynote and its stack of agent launches, Graviton5-powered EC2, Claude Opus 4.8 and a wave of new Bedrock models, plus AgentCore's harness reaching GA and a batch of price cuts. Every item links back to the original announcement.
June 2026 was one of the densest months of the year, anchored by AWS Summit New York City on June 17, where Swami Sivasubramanian's keynote unveiled a full stack of agent launches around a single thesis: agents that compound value over time. Around the Summit, AWS shipped its first Graviton5 EC2 instances, brought Claude Opus 4.8 and a wave of new Bedrock models (OpenAI GPT-5.5/Codex, Gemma 4, Grok 4.3) online, moved AgentCore's harness to GA, and rolled out a batch of price reductions.
Below is a roundup of the month's most notable launches, grouped by theme. Every item links straight to its original announcement.
The Headline: AWS Summit New York City
Swami Sivasubramanian's June 17 keynote grouped its launches into four kinds of agents. Here's the shape of it.
Agents for working — autonomous agents in Amazon Quick amazon-quick
Quick can now create and run multi-step autonomous agents directly in the desktop app, and consolidates email, Slack, calendar, and tasks into a single prioritized activity feed with personalized rules.
Agents for securing — AWS Continuum security
A new AI-native security service that reasons, validates, and acts at machine speed across the full code-vulnerability lifecycle. The AWS Security Agent (now part of Continuum) adds threat modeling, PR code scanning with remediation across major Git platforms, and IDE integrations via Kiro power, Claude Code plugin, and MCP.
Agents for building — Kiro, DevOps Agent & Transform dev-tools
Write, ship, and modernize code in one loop: Kiro gains a native iOS app, AWS DevOps Agent adds release-management capabilities to assess code changes before production, and AWS Transform introduces continuous modernization to reduce tech debt autonomously.
Agents customers create — Bedrock AgentCore harness (GA) agentcore
Go from idea to production-grade agent in minutes. The GA harness handles infrastructure and orchestration, joined by AgentCore Web Search, a Managed Knowledge Base, policy integrations with Guardrails, and the new AWS Context service for mapping organizational data relationships.
Foundation Models on Bedrock
Claude Opus 4.8 on AWS bedrock
Anthropic's most capable GA model, available through both Amazon Bedrock and Claude Platform on AWS. Built for agentic coding, knowledge work, and extended autonomous execution — it sustains longer sessions, recovers from errors, and reads codebases like an engineer, planning before it edits.
OpenAI GPT-5.5, GPT-5.4, and Codex (GA on Bedrock) bedrock
GPT-5.5 and GPT-5.4 are now GA for production workloads on Bedrock, with Codex for AI-powered development via the Codex app, CLI, and IDE integrations (VS Code, JetBrains, Xcode). Pricing matches OpenAI first-party rates and usage counts toward AWS commitments.
Gemma 4 on Amazon Bedrock bedrock
Google DeepMind's Gemma 4 family lands in three variants: a 31B dense model (256K context) for reasoning and coding, a 26B-A4B mixture-of-experts for cost/latency-sensitive work, and a small E2B for low-latency interactive use — all with native function calling, multimodal input, and 35+ languages.
Grok 4.3 from xAI on Amazon Bedrock bedrock
xAI's Grok 4.3 is now available on Bedrock for reasoning, agentic, and enterprise workflows, running on a new price-performance inference engine with tool calling, structured output, and streaming.
Claude Fable 5 on Amazon Bedrock (access later revoked) bedrock
Claude Fable 5 — a Mythos-class model with built-in safeguards — launched on Bedrock on June 9 with extended asynchronous task execution and advanced vision. On June 12, Anthropic asked AWS to revoke access to Fable 5 and Mythos 5 for all users to comply with a US export-control directive; other models, including Opus 4.8, were unaffected. See Anthropic's statement for details.
Agentic AI & Developer Tools
AWS FinOps Agent (preview) finops
An agent for FinOps and engineering teams that answers cost questions, surfaces rightsizing/idle/Savings Plans recommendations from Cost Optimization Hub and Compute Optimizer, opens Jira tickets, and investigates cost anomalies — posting root-cause findings to Slack.
AWS Step Functions — AgentCore agentic reasoning step step-functions
Add AI agent reasoning steps to Step Functions workflows via the managed AgentCore harness — run agents in parallel or sequence, add human approval, and trace every agent decision.
AgentCore Identity — bring-your-own secrets agentcore
Reference existing Secrets Manager secret ARNs in AgentCore Identity Credential Providers, keeping full ownership of secrets governance including custom CMKs, tagging, and automatic rotation.
AWS Blocks — open-source TypeScript framework (preview) dev-tools
Run a fully functional local environment (Postgres, auth, real-time messaging) with no AWS account, then deploy the same code to production AWS services with zero changes — and drop into AWS CDK whenever you need direct resource control.
Kiro Pro Max kiro
A new Kiro tier with higher usage limits, access to the latest frontier models, and additional agentic capabilities for professional developers doing sustained, high-volume work.
Strands Agents — new capabilities strands
The open-source agent toolkit adds better context management in the Harness SDK, an isolated execution environment (Strands Shell), and chaos testing plus red teaming in Strands Evals.
Compute & Serverless
Amazon EC2 M9g and M9gd — first Graviton5 instances ec2
Up to 25% better compute than Graviton4 (35% for web apps and ML inference, 30% for databases). Graviton5 is the first AWS processor with PCIe Gen6 and DDR5-8800 and a 5× larger L3 cache, and the release debuts the Nitro Isolation Engine — making Nitro the first formally verified cloud hypervisor.
Amazon EC2 G7 instances — NVIDIA RTX PRO 4500 Blackwell ec2
AWS is the first major cloud to support NVIDIA RTX PRO 4500 Blackwell Server Edition GPUs. With custom 6th-gen Intel Xeon, G7 delivers up to 4.6× AI inference and 2.1× graphics performance versus G6.
AWS Lambda MicroVMs lambda
A new serverless primitive giving each user or job VM-level isolation with near-instant launch/resume and the ability to suspend and resume for up to 8 hours. Built on Firecracker — ideal for running user- or AI-generated code in multi-tenant apps without trading off isolation, speed, or state.
Amazon ECS — faster service auto scaling ecs
High-resolution (20-second) metrics cut scale-out trigger time from 363s to 86s (76% faster) and total scale-and-provision time from 386s to 109s (72% faster) in AWS benchmarks.
Amazon ECS Managed Instances — Trainium & Inferentia ecs
Create ECS Managed Instances capacity providers with Inferentia2, Trainium1, and Trainium2, and let ECS automatically allocate accelerator resources to your workload.
Amazon EKS supports Kubernetes 1.36 eks
Kubernetes 1.36 promotes User Namespaces to GA and adds Mutating Admission Policies, in-place pod-level vertical scaling, and resource health status reporting — across EKS and EKS Distro.
Storage, Data & Analytics
Amazon S3 annotations s3
Attach up to 1 GB of rich, mutable, queryable context directly to S3 objects — purpose-built for AI agents and autonomous workflows that need to discover, understand, and act on data at scale without a separate metadata system.
Next-gen Amazon OpenSearch Serverless opensearch
A fully managed search and vector engine for agentic AI: scales from zero to thousands of requests/sec (~20× faster than before), up to 60% cheaper than peak-provisioned clusters, with GPU acceleration and new SEARCH and VECTORSEARCH collection types.
Amazon Cognito — multi-Region replication cognito
Synchronize user and machine identity data — credentials, user-pool config, federation — to a standby Region in near real time, so signed-in users keep working through a primary-Region disruption. Available across 16 Regions.
AWS Cost and Usage Report 2.0 — Athena & Redshift integration billing
CUR 2.0 exports are now delivered in the optimal format for your chosen query engine, with supporting infrastructure templates, table definitions, and data-loading instructions.
Amazon Location Service — transit & intermodal routing location
The Routes API adds Transit and Intermodal travel modes to plan journeys combining public transportation with walking, driving, taxi, and rental segments across 13 Regions.
Security, Networking & Operations
Amazon GuardDuty — AI-powered investigations (preview) guardduty
Automatically analyzes findings to separate true threats from benign activity, examining 90 days of context with knowledge graphs and threat intel, and returns a disposition assessment with confidence scoring, MITRE ATT&CK classification, and recommendations in minutes.
Palo Alto Advanced DNS Security on Route 53 Resolver DNS Firewall (preview) route-53
Enforce Palo Alto Networks DNS threat protections directly on Route 53 DNS Firewall rules — no separate firewalls or VPC changes — by subscribing through the embedded AWS Marketplace widget.
Amazon EC2 AMI Watermarks ec2
Embed custom identifiers in private AMIs that carry forward to every derived AMI across copies, Regions, and account shares — combine with Allowed AMIs and Declarative Policies to restrict launches to approved images, at no cost.
AWS Management Console Private Access console
Reach the AWS Console from VPCs without internet connectivity, so enterprises can manage infrastructure through the console while keeping strict network controls in air-gapped environments.
AWS Workload Credentials Provider security
A lightweight client-side provider that automates deployment of ACM-exported certificates and local caching of Secrets Manager secrets across AWS and non-AWS workloads — handling shrinking certificate lifetimes automatically on Windows and Linux.
AWS Local Zone in Hanoi, Vietnam infrastructure
One of the first APAC Local Zones with Amazon S3 and EBS Local Snapshots support, helping customers meet data-residency requirements by storing and backing up data locally.
Customer Experience & Marketplace
Amazon Connect Customer — Agentic CX designer (preview) connect
A no-code canvas where business teams design and deploy AI-powered self-service across voice and digital, blending agentic and deterministic AI in one governed flow. Includes Live Sync, which drives a customer's web or mobile screen in real time as they speak or type.
AWS Marketplace Storefront (GA) marketplace
Partners can spin up their own branded catalog of solutions and services on their website or app in hours, simplifying how channel partners and ISVs manage their cloud marketplace business.
Price Reductions
Amazon S3 Vectors — up to 80% lower query charges s3
Query charges for large vector indexes drop by up to 80%, lowering costs for similarity search across large-scale AI, RAG, and semantic-search workloads — applied automatically, no app changes.
Amazon GameLift Servers — free network bandwidth gamelift
Network bandwidth in and out of AWS is now free for all gen-6-and-later instance types (On-Demand and Spot, no commitment) — you pay only for instance hours.
The Month in a Paragraph
June 2026 was defined by AWS Summit New York City, where Swami Sivasubramanian's keynote organized a wall of launches into four kinds of agents — for working (autonomous agents in Amazon Quick), securing (the new AWS Continuum service), building (Kiro, DevOps Agent, and Transform in one loop), and the ones customers create, powered by a now-GA Bedrock AgentCore harness with Web Search, Managed Knowledge Base, and the AWS Context service. The model catalog expanded dramatically — Claude Opus 4.8, OpenAI's GPT-5.5/GPT-5.4 and Codex reaching GA, Google's Gemma 4, and xAI's Grok 4.3 all landing on Bedrock — though the Mythos-class Claude Fable 5 was pulled days after launch to comply with a US export-control directive. Underneath the AI story, the fundamentals were strong: the first Graviton5 instances (with Nitro now a formally verified hypervisor), NVIDIA Blackwell G7 GPUs, Lambda MicroVMs, S3 annotations, a next-gen OpenSearch Serverless, and a round of price cuts on S3 Vectors and GameLift. If the spring was about giving agents capabilities, June was about giving them a stage — and the infrastructure to run at scale.