AWS adds OpenAI's GPT-6 Astra to Bedrock for business

Amazon Web Services has made OpenAI's GPT-6 Astra generally available on Amazon Bedrock, adding OpenAI's latest frontier model to AWS's managed artificial intelligence platform.

The model is available through supported Amazon Bedrock application programming interfaces. Customers can also configure ChatGPT Work and Codex to use GPT-6 Astra on Bedrock.

GPT-6 Astra arrives as cloud providers compete to host a growing range of third-party and in-house AI models for business customers. By adding another flagship OpenAI model to Bedrock, AWS is expanding the options for companies that want to build AI services without managing the underlying infrastructure.

The model supports a context window of up to 1 million input tokens, allowing users to submit large sets of material such as codebases, contracts and document collections. It is aimed at demanding business workflows involving reasoning, writing, design, and computer- or browser-based tasks.

OpenAI is also introducing enterprise plugins for ChatGPT Work alongside the Bedrock release. Those plugins extend Astra's browser-use functions across common business applications, AWS said.

For corporate buyers, the announcement also highlights the importance of governance and data handling in AI deployments. Existing Bedrock controls can secure workloads, govern access and audit model invocation activity, and inference data is not used for model training, according to AWS.

Broader rollout

The GPT-6 Astra launch was one of several product updates in AWS's latest weekly roundup of announcements. Amazon Quick's desktop app is also now generally available for macOS and Windows.

The desktop version of Quick can work with local files and remain connected to calendars, email and business applications in the background. Conversations, context and agents stay synchronised across desktop and mobile, AWS said.

Quick agents can also continue running after a user closes their computer. That means a task started on a desktop device can keep running in the background and be updated through the mobile app before the user reviews the result later.

AWS has also increased the function timeout for asynchronous and event source mapping invocations on Lambda Managed Instances to 90 minutes from 15 minutes. Synchronous invocations retain the 15-minute maximum.

The longer runtime is intended for workloads such as data processing, media transcoding, financial calculations, AI inference and batch jobs that require longer continuous execution. The same extension applies to steps inside Lambda durable functions, while asynchronous durable functions can still run for up to a year.

Infrastructure changes

In storage, Amazon EBS Volume Clones can now copy volumes across AWS accounts and re-encrypt them with an AWS Key Management Service key in the target account. That could help organisations that separate production and development environments into different accounts, AWS said.

Under the update, a shared production volume can be copied into an isolated development account in the same Availability Zone. The cross-account copy feature works across all volume types, including unencrypted volumes and those encrypted with customer-managed keys, according to AWS.

AWS also announced the general availability of its second-generation single-rack AWS Outposts system. It is a self-contained 42U rack intended for sites that require low latency, local data processing or data residency and do not have space for a larger setup, AWS said.

According to AWS, a single rack can provide up to 2,688 vCPU and 100 TB of Amazon EBS storage. It supports several x86-based Amazon EC2 instance families, including general-purpose, compute-optimised and memory-optimised options, as well as Outposts accelerated networking instances.

Developer tools

AWS also highlighted a new integration between Amazon OpenSearch Serverless and v0 by Vercel. Users can describe a search or AI application in natural language within v0 and generate a full-stack application backed by OpenSearch Serverless, AWS said.

The setup provisions a collection, indexes data and uses the OpenSearch Serverless endpoint for full-text and vector search in retrieval-augmented generation workloads. Users can provision the service under a new AWS account or connect an existing one.

For software modernisation, AWS Transform for .NET Modernisation is now generally available through a command-line interface. Customers can trigger an AWS-managed .NET modernisation with a single command, then run it interactively or incorporate it into an existing pipeline, AWS said.

The command-line option sits alongside existing access to AWS Transform for .NET through the web application, Visual Studio IDE, Kiro Power and MCP agents. The tooling can be used to upgrade language versions, migrate frameworks, analyse codebases and make custom or out-of-the-box transformations, according to AWS.

AI workflow debate

Beyond product launches, AWS drew attention to an article by Clare Liguori, Senior Principal Engineer at AWS, on what she called “frontier engineering”. The piece argues that the biggest productivity gains from AI coding assistants come not simply from using different tools, but from changing how software teams organise their work around agents.

AWS also highlighted internal and external research on AI trust and application development. That included a security benchmarking effort called Deception Benchmark, which tests whether models can distinguish genuine vulnerabilities from code that appears risky but is safe, and an open-source Nx Plugin for AWS designed to generate application building blocks and related AWS infrastructure.

The roundup underscored the breadth of AWS's current AI push, spanning foundation models, software agents, developer tooling, cloud infrastructure and security testing. At the centre was the addition of GPT-6 Astra to Bedrock, extending AWS's catalogue of models available to business customers.