What did AWS announce about OpenAI and Amazon Bedrock?

On April 28, 2026, at the What's Next with AWS event, AWS and OpenAI extended their partnership: OpenAI's GPT-5.5 and GPT-5.4 models are coming to Amazon Bedrock in limited preview, alongside Codex on Amazon Bedrock (a coding agent) and Amazon Bedrock Managed Agents powered by OpenAI.

Is GPT-5.5 on Bedrock already available for production?

Not yet. AWS describes it as a limited preview, meaning controlled access to evaluate the model ahead of general availability. It is worth exploring for test cases and architecture work, while waiting for general availability before committing critical workloads.

Why use GPT-5.5 through Bedrock instead of directly with OpenAI?

Because you access the model through the same APIs, security controls, governance and billing that Bedrock already provides across the rest of your AWS architecture. There is no need to stand up a separate integration or a new security model to add another model provider.

What is Amazon Bedrock AgentCore and how does it relate to this announcement?

AgentCore is AWS's service for taking AI agents from prototype to production with infrastructure, observability and controls ready to operate at scale. On June 17, 2026 it added new capabilities (a generally available harness, managed web search, a managed Knowledge Base), and it is the same environment where agents using OpenAI models via Bedrock will run.

What should my company evaluate before adopting this?

A concrete use case (what process it improves, what the agent decides or executes), the governance and sensitive-data requirements of your industry, and whether to start with a scoped pilot in preview or wait for general availability. It is not an all-or-nothing decision — you can evaluate it without committing production workloads.

GPT-5.5 on Amazon Bedrock: what it means

On April 28, 2026, at the What’s Next with AWS event, AWS and OpenAI announced an expanded partnership: OpenAI’s latest models are coming to Amazon Bedrock. For a company already running on AWS, this is not just a catalog update — it changes how you can access models from different providers without leaving the security and governance environment you already know.

What AWS announced with OpenAI

At the event, Matt Garman (CEO of AWS), Colleen Aubrey (SVP of Amazon Applied AI Solutions), Julia White (CMO of AWS) and OpenAI leaders presented three announcements:

OpenAI models on Amazon Bedrock (limited preview): OpenAI’s latest models, including GPT-5.5 and GPT-5.4, are available in preview on Amazon Bedrock. You access them through the same Bedrock APIs you already use, with unified security, governance and cost control.
Codex on Amazon Bedrock (limited preview): OpenAI’s coding agent, available within the AWS environments teams already operate in, authenticated with AWS credentials.
Amazon Bedrock Managed Agents, powered by OpenAI (limited preview): a managed way to build production-ready agents that combine OpenAI’s models with AWS infrastructure.

GPT-5.5 and GPT-5.4 on Amazon Bedrock: what it means

The core of the announcement is not just “one more model available” — it is where you access it. Before this, adding an OpenAI model to an AWS architecture meant a separate integration, with its own authentication, billing and security controls. With GPT-5.5 and GPT-5.4 on Bedrock, that model is consumed through the same API, IAM and observability layer already used across the rest of your generative AI platform on Amazon Bedrock.

For an IT team, that removes friction from a decision that used to be harder: choosing the best model for each task without multiplying the risk surface. Bedrock already lets you compare and combine foundation models from different providers; adding OpenAI extends that ability to pick the right tool for each case —writing, analysis, code, agents— without leaving AWS’s security perimeter.

Codex and the Managed Agents powered by OpenAI

Alongside the language models, AWS and OpenAI brought two more execution-focused pieces to Bedrock: Codex, OpenAI’s coding agent, now accessible within the AWS environment development teams already work in; and Bedrock Managed Agents, a way to build production agents that combine OpenAI’s frontier models with AWS’s managed infrastructure, without assembling that orchestration from scratch.

This approach —model plus agent plus managed infrastructure— matches where the industry is heading: from assistants that answer questions to AI agents that execute multi-step tasks.

Bedrock keeps adding agent capabilities

April’s announcement was not the only recent move in Bedrock. On June 17, 2026, at the AWS Summit in New York, AWS presented new capabilities for Amazon Bedrock AgentCore —the service for taking AI agents from prototype to production—: the AgentCore harness reached general availability, a managed Knowledge Base was added for enterprise RAG, and a managed web search tool lets agents ground their answers in current, verifiable information. You can read more about how AgentCore works in our guide Bedrock AgentCore: from prototype to production.

Read together, both announcements send a clear signal: AWS is investing both in which models you can use and in the infrastructure to run agents with them securely and observably.

What it means for your company today

A model being in preview is not a signal to move immediately, nor to ignore it. It is the moment to:

Identify a scoped use case worth evaluating GPT-5.5 or GPT-5.4 for —not replace your entire AI architecture at once.
Review your industry’s data governance requirements (banking, insurance, healthcare, retail) against what the preview offers.
Take advantage of the fact that evaluation does not require a new integration —if you already use Bedrock, testing an additional model is a configuration change, not an infrastructure project.
Wait for general availability before committing critical production workloads.

How Caleidos helps

At Caleidos we help companies already running on AWS evaluate, with sound judgment, which generative AI capabilities add real value to their operation, without betting their entire architecture on a preview announcement. Our Agentic AI on AWS practice helps define the use case, the governance controls and the path from pilot to production on Amazon Bedrock.

Frequently asked questions

What did AWS announce about OpenAI and Bedrock? That OpenAI’s GPT-5.5 and GPT-5.4 models, along with Codex and Managed Agents, are coming to Amazon Bedrock in limited preview (April 28, 2026).

Is it already available for production? No, it is a limited preview. It is worth evaluating with a scoped pilot and waiting for general availability for critical workloads.

Why use Bedrock instead of OpenAI directly? Because you keep the same APIs, security, governance and billing already used across the rest of your AWS architecture.

Want to evaluate which model fits your use case?

Let’s talk about your generative AI strategy and we’ll help you define the path from pilot to production on Amazon Bedrock.

GPT-5.5 and GPT-5.4 arrive on Amazon Bedrock: what AWS announced and what it means for your business