OpenAI's GPT-5.5, GPT-5.4 and Codex Land on Amazon Bedrock

Amazon Web Services on June 1, 2026 announced that OpenAI’s top models — GPT‑5.5, GPT‑5.4 and the Codex coding agent — are now generally available on Amazon Bedrock.

The release lets Bedrock customers call OpenAI’s frontier models from within AWS infrastructure rather than via a separate OpenAI account, bringing the models into the Bedrock product catalog and supported APIs.

AWS says the OpenAI models on Bedrock inherit the cloud provider’s enterprise controls — including AWS Identity and Access Management, network isolation via PrivateLink, encryption, guardrails and CloudTrail logging — so teams can run inference inside their existing governance and audit frameworks.

The Bedrock general availability follows a limited preview that began in late April, part of an expanded partnership between AWS and OpenAI that ended the practical exclusivity OpenAI earlier had with a single cloud provider. The April preview showcased Managed Agents and initial integration paths ahead of the June GA.

For enterprises, the change removes one common procurement friction: IT and security teams that already approve AWS infrastructure can now access OpenAI’s frontier models without an extra vendor approval process or separate cloud contracts. Customers and partners told industry outlets that having the models inside Bedrock simplifies compliance and contracting.

Amazon positions Bedrock as a multi-model platform that now combines models from several providers alongside OpenAI’s frontier family, letting teams choose by capability, cost and governance needs rather than by which cloud hosts a model. That consolidation is part of AWS’s broader pitch to make Bedrock the model hub for enterprise AI.

OpenAI and AWS also highlighted Codex’s developer tooling: Codex on Bedrock is offered as an agent for writing, reviewing, debugging and modernizing code, with integrations for CLI and IDE workflows. OpenAI said Codex already supports millions of developer users weekly and that the Bedrock integration routes inference through Bedrock’s APIs.

Technically, AWS instructs developers to call the OpenAI models through Bedrock’s Responses API and provides code samples and configuration tips in a 'Get started' blog and the Bedrock docs. AWS noted that, at GA, the new OpenAI endpoints are available programmatically and that console-level UI support would follow.

Region availability and pricing are set at launch: AWS’s blog lists initial regions for each model and describes token-based pricing with no per-developer seat fees. Customers with data residency needs can keep processing within the chosen Bedrock region, AWS said.

The move also reshaped multi-cloud distribution dynamics. Analysts and news outlets framed the AWS integration as a key step toward a true multi-cloud market for frontier models, after OpenAI relaxed earlier exclusivity arrangements that had tied it more tightly to one hyperscaler. That shift means enterprises can now weigh model choice across providers while keeping workloads governed inside the cloud they already use.

Partners and systems integrators said the availability makes it easier to build production systems that combine multiple models and data sources behind corporate controls. But they also warned that operational complexity — capacity planning, cost controls, and model orchestration — will move from vendor setup to enterprise architecture decisions.

For AWS customers the practical next steps are straightforward: review the Bedrock model cards, test the Responses API in a staging region, and update IAM and VPC settings to reflect any new inflection points where model inference touches sensitive data. AWS and OpenAI published docs, examples and a getting-started guide to help teams migrate or build new applications on Bedrock.