← back to blog
cloudFeatured·Aug 1, 2026·12 min read

What's New with AWS — July 2026: The Month in Review

A themed roundup of every headline launch AWS shipped in July 2026 — back-to-back Anthropic model drops (Claude Sonnet 5 then Claude Opus 5), OpenAI's GPT-5.6 family on Bedrock, Graviton5 compute-optimized instances, up to 60% lower GPU management fees, and a wave of serverless and security releases. Every item links back to the original announcement.

SJ
Sabin Joshi
DevOps Engineer
#aws#whats-new#bedrock#claude#graviton5#serverless#roundup

After June's AWS Summit fireworks, July 2026 settled into a steadier — but no less busy — rhythm. There was no single mega-event this month; instead, the story was a remarkable Anthropic model cadence on AWS, with Claude Sonnet 5 landing early in the month and the flagship Claude Opus 5 arriving by month's end. Around that, AWS shipped OpenAI's GPT-5.6 family on Bedrock, its first Graviton5 compute-optimized instances, a meaningful GPU cost cut, and a deep bench of serverless, security, and developer-tooling launches.

Below is a roundup of the month's most notable launches, grouped by theme. Every item links straight to its original announcement.

ℹ️
A note on scope. AWS pushes far more to its What's New feed than any single post can capture — often dozens of items a week, many of them regional expansions or incremental availability updates. This covers the headline, developer-relevant launches AWS itself highlighted in its weekly roundups for the month. For the exhaustive, item-by-item list, the canonical source is the What's New with AWS page.
4
weekly roundups
2
new Claude models on AWS
60%
lower GPU management fees
2
new Local Zones

The Headline: Anthropic's Model Cadence on AWS

July bookended the month with two Claude launches — a capable new Sonnet at the start and the flagship Opus at the end.

Claude Opus 5 on AWS bedrock

Anthropic's most advanced Opus model yet, matching Claude Fable 5's top-tier intelligence in many domains at Opus-tier pricing. On Amazon Bedrock it ships with zero data retention (ZDR) enabled by default, and it's available through both Bedrock and Claude Platform on AWS.

OpenAI GPT-5.6 — Sol, Terra & Luna on Bedrock bedrock

OpenAI's smartest family yet arrives on Bedrock's next-gen inference engine, spanning capability tiers from flagship reasoning (Sol) to balanced performance (Terra) to fast, cost-efficient inference (Luna) — all through the Responses API.

Agentic AI & Model Tooling

Claude apps gateway for AWS claude

A self-hosted control plane giving organizations centralized control over access, cost, and policy for Claude Code and Claude Desktop. It connects to any OIDC identity provider, enforces managed settings on every request, routes inference to Bedrock or Claude Platform on AWS, and supports per-user and per-group spend caps.

Loom for AWS (open source) open-source

An open-source enterprise platform for building agents with Strands Agents and deploying them on Bedrock AgentCore Runtime — with a unified management UI, identity-provider integration, RBAC/ABAC multi-tenant security, automated cost-attribution tagging, and AWS Agent Registry integration.

Bedrock AgentCore — unified observability agentcore

Agent traces and prompts now land in the same CloudWatch log group as your agent's logs, so debugging a single invocation happens in one place — with fine-grained access control and CMK encryption at the individual agent level.

SageMaker Studio — one-click Hugging Face deployment sagemaker

Go from discovering a model on Hugging Face to customizing or deploying it in SageMaker Studio in a single click. New customers get a Studio environment in seconds; verified customers get default GPU access to G5, G6, and G4dn without quota requests.

SageMaker AI — faster inference scale-out sagemaker

Container image caching enables up to 2× faster end-to-end scaling for generative AI models during scale-out events.

Amazon CloudWatch — coding agent insights cloudwatch

Gives engineering leaders visibility into how AI coding tools drive value. It integrates with the Claude apps gateway to collect telemetry from Claude Code without extra instrumentation, and also supports Codex and GitHub Copilot — metrics built on OpenTelemetry.

Serverless & Developer Experience

One-click Lambda setup prompt for coding agents lambda

A prompt that configures your coding agent with AWS Serverless skills and the Serverless MCP server, embedding best practices from the start — with install support for Claude Code, Kiro, Cursor, GitHub Copilot, Codex, Devin Desktop, and OpenCode.

AWS Lambda durable execution SDK for .NET (GA) lambda

Build resilient, long-running workflows in C# with Lambda durable functions — no custom progress tracking or external orchestration. It checkpoints automatically and can pause for up to a year, ideal for payment pipelines, agent orchestration, and human-in-the-loop approvals.

Self-managed code storage on AWS Lambda lambda

Reference source code directly from your own S3 buckets without Lambda creating intermediate copies — removing code storage limits and cutting activation time after creates and updates.

AWS CloudFormation Express mode cloudformation

Up to 4× faster infrastructure deployment, so AI agents and developers get deployment confirmation in seconds and iterate faster — available in all commercial Regions at no additional cost.

Amazon EKS — Kubernetes version rollbacks eks

Reverse a cluster upgrade within seven days with no cluster rebuilds — turning Kubernetes version upgrades into a reversible, low-risk operation.

Amazon SQS turns 20 sqs

Two decades on from its July 2006 public launch, SQS's core function — decoupling producers from consumers — remains the reason customers reach for it. A nice milestone for one of AWS's foundational services.

Compute & Infrastructure

Amazon EC2 C9g and C9gd — Graviton5 compute-optimized ec2

Up to 25% better compute performance than Graviton4 instances, a 5× larger cache, the fastest memory of any cloud processor instances, and local NVMe storage options (C9gd).

EKS Auto Mode & ECS Managed Instances — GPU fees down up to 60% containers

From July 1, management fees for accelerated instances drop: G-series down 35%, and P-series and AWS Trainium down 60%. The reductions apply automatically to existing clusters, no action required.

AWS Local Zone in Athens, Greece infrastructure

The second Local Zone in EMEA with Amazon S3 and EBS Local Snapshots support, letting you process and store data in-country (EC2 C7i/M7i/R7i, S3 One Zone-IA, EBS, ECS) with single-digit millisecond latency.

Storage, Data & Analytics

Amazon S3 — same-day transitions to Standard-IA and One Zone-IA s3

Transition objects to S3 Standard-IA and One Zone-IA the same day they're created, dropping the previous 30-day minimum — up to 40% lower storage cost for backups, log analytics, and compliance data that goes cold within hours.

Amazon OpenSearch Service — optimized for log analytics opensearch

A new engine purpose-built for log analytics delivers up to 4× better price-performance on internal benchmarks while keeping full-text search — aggregations and precise text search in one place.

Amazon Aurora DSQL change data capture (GA) aurora

Stream insert/update/delete change events to Kinesis Data Streams to sync microservices, trigger Lambda, or deliver changes to S3, Redshift, and OpenSearch via Data Firehose — with zero impact on database performance and no infrastructure to manage.

SageMaker Unified Studio — OpenSearch support sagemaker

Query and analyze OpenSearch search and log data alongside Redshift, S3, and relational sources in one governed environment — useful for correlating operational search data with transactional data.

Open-source Bulk Executor for Amazon DynamoDB dynamodb

Run bulk operations across all items in a table — count, find, delete, update — with no custom code, even at large scale.

Amazon CloudWatch — alarms from log queries cloudwatch

Create alarms directly on log query results and set thresholds in a single workflow, without first building metric filters or custom metrics.

Security & Identity

AWS Security Hub — Network Scanning security-hub

Probes your resources from the internet to detect actual reachability — discovering public IPs, VMs, and load balancers across AWS and Azure, identifying reachable ports and the services behind them, and correlating findings for broader risk. Included with Security Hub Essentials at no extra cost.

Security Hub extends to Microsoft Azure security-hub

Unified posture, vulnerability management, and response across both clouds — auto-discovering Azure VMs, container images, Function Apps, and identities, with AWS and Azure findings in the same prioritized view.

OAuth support for AWS MCP Server mcp

Connect agents to the AWS MCP Server using browser-based OAuth with the same credentials you use for the Console or CLI — supporting IAM federation, IAM Identity Center, and short-lived tokens with automatic management, plus a non-interactive flow for headless use.

ACM — ACME protocol support acm

AWS Certificate Manager now supports the ACME protocol, automating issuance and renewal of public TLS certificates with standard, widely adopted tooling.

Amazon Cognito — import users with password hashes cognito

Include password hashes in CSV user imports so migrated users can sign in immediately with existing credentials, instead of resetting on first sign-in.

Customer Experience

Amazon Connect — more natural agentic voice connect

Agentic voice experiences across 50+ languages with 100+ new voice options, adapting to a caller's tone and sentiment across voice and digital channels — so AI interactions sound more fluid in far more of the languages customers actually speak.

Amazon WorkSpaces for AI agents (GA) workspaces

AI agents can now securely access and operate desktop applications through managed WorkSpaces environments — no application modernization or custom integrations required.

⚠️
A note on the model timeline: Claude Opus 5 (July 27) explicitly positions itself against Claude Fable 5, matching much of its intelligence at Opus-tier pricing with zero data retention on by default. Fable 5 itself had launched on Bedrock back in June before access was revoked under a US export-control directive — so Opus 5 effectively became the practical way to get near-Fable-5 capability on AWS this month. Also note the quarterly AWS Service Availability update landed July 6, moving several services (including Amazon Bedrock Agents, now "Agents Classic," Kendra, and Q Business) into maintenance.

The Month in a Paragraph

July 2026 was a model-cadence month for AWS. Where June had a Summit to anchor it, July's throughline was Anthropic: Claude Sonnet 5 opened the month and the flagship Claude Opus 5 closed it — the latter matching much of Fable 5's intelligence at Opus pricing with zero data retention by default — while OpenAI's GPT-5.6 (Sol, Terra, Luna) filled out the Bedrock catalog. The tooling around those models matured in step, with the self-hosted Claude apps gateway, the open-source Loom platform, unified AgentCore observability, and CloudWatch coding-agent insights. Underneath the AI layer, AWS kept shipping the fundamentals: the first Graviton5 compute-optimized C9g instances, a welcome GPU management-fee cut of up to 60%, Lambda durable execution for .NET, same-day S3-IA transitions, a log-optimized OpenSearch engine, Aurora DSQL CDC, and a broader Security Hub that now scans from the internet and reaches into Azure. No single blockbuster — just a deep, steady month that pushed both the frontier and the plumbing forward.