Stunning OpenAI on AWS Bedrock: A Powerful 2026 Guide

Indeed, on April 28, 2026, indeed, OpenAI on AWS Bedrock went live in limited preview. The move ended nearly seven years of Microsoft Azure lock-in over OpenAI's frontier models in a single joint launch from AWS and OpenAI.

In fact. So, if you build on AWS and you've been waiting to bring GPT models into your existing setup without juggling two cloud bills, your stack just got simpler. OpenAI on AWS Bedrock changes how teams ship AI features. You can now route API calls through the same VPC, IAM roles, and log tooling you already use for every other Bedrock workload.

Of course, here's what shipped, what it costs, and how to decide if it fits your roadmap.

Key Takeaways

OpenAI on AWS Bedrock launched April 28, 2026 — GPT-5.5 and GPT-5.4 are live in limited preview alongside Codex and Managed Agents.
Codex reaches a wider audience — over 4 million developers already use it weekly, and now they can access it through their AWS account.
Managed Agents handle the hard parts — rollout, tool use, glue, and governance ship with native AWS security and audit controls.
Existing AWS deals cover OpenAI usage — Codex and OpenAI model spend can apply toward your AWS cloud agreements.

What Is OpenAI on AWS Bedrock?
OpenAI on AWS Bedrock: live Models
Codex with OpenAI on AWS Bedrock
Managed Agents on OpenAI on AWS Bedrock
OpenAI on AWS Bedrock Pricing and Access
OpenAI on AWS Bedrock vs Microsoft Azure
Who Should Use OpenAI on AWS Bedrock?
How to Get Started With OpenAI on AWS Bedrock

What Is OpenAI on AWS Bedrock?

Hence, openAI on AWS Bedrock is the official link that brings OpenAI's frontier models, the Codex coding agent,. Plus, managed Agents into Amazon's managed API calls tool. In fact, all three tools launched in limited preview on April 28, 2026. Plus, with launch promised within weeks.

Thus, before this launch, AWS customers who wanted GPT had to call OpenAI's API directly or pipe traffic through Azure. That meant separate billing, separate authentication, and separate audit reviews. Now, the same Bedrock APIs that already serve Anthropic, Meta, Mistral, Cohere, and AI21 models also serve OpenAI.

In short, according to AWS, Amazon Bedrock powers generative AI for more than 100,000 organizations worldwide. Adding OpenAI to that catalog turns Bedrock into a true model marketplace rather than a curated alternative to OpenAI.

Besides, for developers, the useful change is small but meaningful. Of course, you authenticate with AWS credentials, your tokens flow through Bedrock setup,. Plus, audit logs land in the same place as every other AWS tools. You're not signing up anywhere new.

For example, have you been managing two API keys for AI work? You can probably collapse that down now.

OpenAI on AWS Bedrock: live Models

In contrast, the launch headline is GPT-5.5, OpenAI's newest frontier model. Plus, with GPT-5.4 live alongside it. Indeed, both are accessible through standard Bedrock InvokeModel calls and through the agentic Bedrock APIs.

As a result, the model lineup is structured to match how teams actually deploy:

GPT-5.5 — top-of-stack reasoning, coding, and long-context work
GPT-5.4 — slightly older but cheaper for high-volume tasks
Codex (model and agent) — built for code engineering tasks

Above all, in fact, the Bedrock setup works with streaming responses, tool calls,. Plus, structured output — the same primitives Bedrock already exposes for other model vendor. If you've written Bedrock code for Claude or Llama, switching the model ID is the only meaningful change.

Bedrock now hosts nearly 100 serverless models across vendor. Plus, with 4.7x adoption growth tracked over the past year, according to AWS partner data published this April. Hence, openAI joining that catalog meaningfully shifts where token spend will land.

Codex with OpenAI on AWS Bedrock

So, codex is OpenAI's self-run coding agent. It writes, refactors, explains, and tests code,. So, and on Bedrock it gets first-class access to AWS systems where most teams already build. Plus, ship.

Yet, according to OpenAI, more than 4 million people use Codex every week. Thus, bringing that user base into AWS is a real distribution play. So, Codex now slots into CI pipelines, version control hooks, and IDE plugins without leaving the AWS edges.

Codex access points on Bedrock

Codex CLI — script Codex from any terminal authenticated with AWS credentials
Desktop app — chat-based development flows on macOS and Windows
VS Code add-on — inline edits, completions, and agent runs from the editor

Still, for DevOps teams already using the AWS DevOps Agent, Codex gives a second AI surface that is specifically tuned for code rather than setup. Both tools can sit in the same pipeline and fits with each other.

Often, one flow that suddenly becomes simple: a Codex agent reads a CloudWatch alarm, generates the fix, opens a pull request,. Plus, runs tests — all without leaving the AWS account it lives in.

Managed Agents on OpenAI on AWS Bedrock

Also, managed Agents are the third tools in the launch. In short, they sit on top of OpenAI models. So, and handle the messy parts of running an agent in live: tool use, multi-step glue, persistence,. Plus, governance.

Indeed, if you've built an agent from scratch, you know the boilerplate. You write the loop, manage the tool registry, handle retries, and bolt on permissions. Besides, managed Agents replace that scaffolding with an AWS-native runtime that has security and audit controls baked in.

Bedrock AgentCore, the SDK behind agent workloads, has been downloaded over 2 million times in five months since its preview shipped, per AWS partner reporting. That tells you the appetite for managed agent runtimes is large and growing fast.

In fact, for a deeper look at how this category is evolving, the Agentic AI guide on this site walks through the architecture patterns Managed Agents implement.

What Managed Agents handle for you

Tool definitions and call routing
Multi-turn state and persistence
Cost and concurrency limits
IAM-scoped permissions for downstream AWS resources
Audit logging through CloudTrail

Codex with OpenAI on AWS Bedrock OpenAI coding agent inside AWS systems

OpenAI on AWS Bedrock Pricing and Access

Of course, aWS has not published a public per-token price list for OpenAI models on Bedrock at launch. For example, the official line is that full pricing arrives with launch in the coming weeks.

Hence, indeed, two things are confirmed:

Codex spend applies toward existing AWS cloud deals. This adds large firm Discount Program agreements.
In short, usage billing flows through the standard Bedrock bill. So, no separate OpenAI subscription is required.

Thus, in short, for finance teams, that is the headline. Your AI spend can roll into the deal you have already agreed on. If your organization commits to a multi-year AWS deal, your OpenAI usage helps you hit that deals not splitting your cloud spend.

In short, above all, to request preview access, you send a form through the Bedrock console, the same way you would request access to any other gated foundation model. AWS is approving accounts in waves while the preview ramps.

OpenAI on AWS Bedrock vs Microsoft Azure

Besides, the Microsoft side of this story is nuanced. Azure does not lose OpenAI. So, but it does lose lock-in. As a result, according to Microsoft, Azure remains OpenAI's primary cloud partner under a license that runs through 2032.

For example, the exact wording from the agreement: OpenAI products will ship on Azure first, unless Microsoft cannot or chooses not to works with the necessary features. That gives Azure a structural advantage on time-to-market while removing the legal block on Bedrock distribution.

In contrast, microsoft also continues to receive 20% of OpenAI revenue through 2030, though that figure is now subject to a cap whose value the companies have not disclosed.

As a result, for your existing Azure OpenAI rollout, nothing breaks. So, endpoints stay live, SLAs stay in force, and feature releases land on Azure first. The change matters mainly when you architect a new project. For new workloads, the choice between Azure OpenAI and OpenAI on AWS Bedrock is now a real procurement decision rather than a forced one.

Quick side-by-side

Time to new model release: Azure usually first, Bedrock follows
Existing AWS deals apply: Yes on Bedrock, no on Azure OpenAI
Native link with AWS tools: Bedrock wins
Native link with Microsoft 365 and Copilot ecosystem: Azure wins

Who Should Use OpenAI on AWS Bedrock?

Above all, the link is most valuable for three groups. Yet, if you fit one of these, the migration math is easy.

1. AWS-native big firms

So, if your organization already runs the bulk of compute on AWS, pulling OpenAI into Bedrock removes a multi-cloud tax. You consolidate billing, IAM, log, and audit reviews under one vendor. Still, procurement teams will appreciate the simpler vendor footprint.

2. Engineering teams that want Codex without leaving AWS

Yet, for development organizations standardized on AWS tooling, Codex on Bedrock gives you the most-used coding agent inside the same authentication boundary as your code, your CI, and your secrets manager. You stop pasting API keys into developer machines.

3. Often, audited industries with AWS-only data edges

Still, healthcare, financial tools, and government workloads often require all data to remain inside an AWS system governed by specific audit attestations. Routing OpenAI calls through Bedrock keeps the trace inside that edges and inherits AWS's audit posture. This adds HIPAA-set tools status.

Often, sound familiar? If your last security review pushed back on a separate OpenAI vendor, that conversation just got easier.

How to Get Started With OpenAI on AWS Bedrock

Also, the setup path mirrors any other Bedrock model. Also, here is the order most teams will follow.

Request preview access through the Bedrock console under Model access. Look for OpenAI in the vendor list.
Update IAM with the bedrock:InvokeModel permission scoped to the OpenAI model ARNs.
Pick an SDK pattern. The boto3 invoke_model call works the same as it does for other Bedrock vendor.
Enable CloudTrail for the Bedrock API to capture audit logs at the model-call level.
For Codex, install the CLI or VS Code add-on and authenticate using your AWS credentials.

Indeed, one useful tip from my real use setting up early Bedrock previews: build a thin wrapper around the model call so you can swap vendor without rewriting apps code. In fact, the same pattern saved teams months of work when Anthropic. So, and Meta dropped new model versions on Bedrock.

In fact, also, if you are new to Bedrock entirely, the wider tool is worth understanding. The Google Cloud AI partnership guide on this site offers a useful side-by-side of how the three big vendor structure their AI marketplaces.

Summary

Of course, openAI on AWS Bedrock is the cleanest path for AWS-native teams to use GPT-5.5, GPT-5.4, Codex,. Plus, managed Agents inside a familiar account boundary. Of course, codex spend applies to existing AWS deals, IAM and CloudTrail handle access out of the box. The link removes the extra work of running a second cloud relationship for AI alone. For new projects, this is now a real evaluation rather than a forced Azure choice.

Frequently Asked Questions

Is OpenAI on AWS Bedrock live?

Hence, no. Indeed, as of April 28, 2026, all three tools — OpenAI models, Codex,. So, and Managed Agents — launched in limited preview. AWS has stated that launch is set within the following weeks. You request access through the Model access page in the Bedrock console.

Which models ship with OpenAI on AWS Bedrock?

Thus, the launch wave adds GPT-5.5 and GPT-5.4. Hence, codex is live as a separate tools,. So, and Managed Agents wrap OpenAI models with an AWS-native agent runtime. AWS has not yet committed to a fixed schedule for adding additional OpenAI models to Bedrock.

Does OpenAI on AWS Bedrock cost extra on top of my AWS bill?

In short, no. API calls billing flows through the standard Bedrock bill and does not require a separate OpenAI subscription. Thus, codex usage applies toward existing AWS cloud deals. This adds large firm Discount Program agreements.

Will my Azure OpenAI rollout break because of this launch?

Besides, no. Azure remains OpenAI's primary cloud partner through 2032, all existing endpoints continue to operate,. So, and OpenAI products still ship on Azure first. In short, the April 28, 2026 launch only ends lock-in, not the partnership itself.

Can Codex on Bedrock replace existing AWS coding tools?

For example, of course, it depends on your flow. Codex is for on code writing and review. AWS-native tools like the AWS DevOps Agent focus on setup and work. Besides, in fact, most teams will use both. So, Codex for code making, the DevOps Agent for system auto-runs. The two surfaces fits with each other inside a single AWS account.

Editorial Disclosure: This article was researched and drafted with AI assistance. Then, reviewed, fact-checked, and edited by Bhanu Prakash to ensure accuracy and provide hands-on insights from real-world real use.

About the writer

Bhanu Prakash is a cybersecurity and cloud computing professional with hands-on real use in AWS, Bedrock, and AI agent flows. For example, he shares useful guides and career advice at ElevateWithB.

What to Read Next: Check out the deep dive on Agentic AI in 2026 to see how Managed Agents fit the wider self-run-agent landscape.