OpenAI announced on 28 April 2026 that its models, including Codex and a new "Managed Agents" offering, are now available directly through Amazon Web Services [source]. The move marks a significant shift in OpenAI's distribution strategy, which has historically centered on its own API and a partnership with Microsoft Azure.

According to the announcement, AWS customers can now access GPT-4, GPT-3.5, and Codex models through AWS Bedrock and other AWS services. The "Managed Agents" product is described as a hosted service for building autonomous AI workflows, though technical specifications were not detailed in the initial release.

The announcement did not clarify whether models deployed on AWS will share the same training data, fine-tuning, or safety filters as those on OpenAI's native platform or Azure. No information was provided on version parity, update cadence, or whether AWS-hosted models will exhibit identical behavior to their counterparts elsewhere.

OpenAI's blog post emphasized enterprise integration benefits, including AWS IAM compatibility and VPC deployment options. Pricing details were not disclosed, nor was any comparison made to existing OpenAI API rates.

The timing follows recent enterprise demand for multi-cloud AI deployments and comes amid ongoing scrutiny of exclusive cloud partnerships in the AI sector. AWS has been expanding its generative AI offerings through Bedrock, which already hosts models from Anthropic, Stability AI, and others.

Users relying on OpenAI's existing API or Azure OpenAI Service should note that this expansion introduces a third deployment environment. No guidance was provided on how to ensure consistent model behavior across platforms or whether cross-platform testing is recommended for production workloads.