GPT-5.5, GPT-5.4, and Codex Are Now Generally Available on Amazon Bedrock

GPT-5.5, GPT-5.4, and Codex became generally available on Amazon Bedrock on June 1, 2026. For the millions of enterprises that already run their infrastructure on AWS, this removes a meaningful obstacle: getting OpenAI’s most capable models into production without stepping outside the governance and compliance stack they already have in place.

What’s Actually Available

There are two distinct offerings here, and it’s worth being clear about what each one does.

OpenAI on Bedrock gives teams access to GPT-5.5 and GPT-5.4 for building AI applications through AWS-native controls. GPT-5.5 is positioned for the most demanding workloads, with strong reasoning, coding, and agentic capabilities. GPT-5.4 sits slightly behind it on raw capability but offers better price-performance and is available across two regions: US East (Ohio) and US West (Oregon). GPT-5.5 is currently US East (Ohio) only.

Codex on Amazon Bedrock brings OpenAI’s software engineering agent into the AWS environment. Codex can hold context across entire repositories, reason through ambiguous failures, check assumptions using tools, and carry changes through surrounding code with full awareness of how systems connect. It’s available through the Codex App, Codex CLI, and IDE integrations for Visual Studio Code, JetBrains, and Xcode, with all model inference routed through Bedrock.

The Security and Compliance Story

This is where the announcement becomes genuinely significant for regulated industries.

Every API call to these models inherits AWS’s existing governance controls: IAM permissions, VPC and PrivateLink isolation, KMS encryption, and CloudTrail audit logging. Prompts and responses are not used to train models and are not shared with OpenAI. Inference stays within the selected AWS Region, which matters for data residency requirements. Both Commercial and GovCloud regions are supported.

For a bank, hospital system, or government contractor, these are the controls that determine whether something can actually move from a proof of concept into production. The models being good is table stakes. Having them sit inside the same perimeter as everything else is what closes the deal with legal and security teams.

Codex Billing That Works With Existing AWS Spend

Codex on Bedrock uses pay-per-token pricing with no seat licenses and no per-developer commitments. Crucially, usage counts toward existing AWS commitments. If your organisation has an Enterprise Discount Program agreement or reserved capacity commitments with AWS, Codex usage rolls into that. This makes procurement straightforward in a way that a separate OpenAI contract often isn’t.

GPT-5.5 and GPT-5.4 pricing matches OpenAI’s standard rates with no additional Bedrock markup.

How to Access It Technically

If you want to call GPT-5.5 via the API, the base URL is https://bedrock-mantle.us-east-2.api.aws/openai/v1 with the model ID openai.gpt-5.5. The models are accessible through the OpenAI Responses API using the OpenAI SDK or curl, so existing OpenAI integrations require minimal changes beyond the endpoint and authentication method.

For Codex specifically, there’s a built-in amazon-bedrock model provider. You set model_provider = "amazon-bedrock" and specify the Bedrock model ID directly, with support for nested AWS profile and region overrides.

During high-traffic periods, Bedrock queues requests rather than rejecting them, and hardware-level progress tracking allows requests to resume from node failures rather than starting over. For long-running Codex tasks, that resilience matters.

Who’s Already Using It

Amgen’s CTO Sean Bruich called out GPT-5.5 specifically, noting that frontier models offer meaningful advances in capability, quality, and consistency for a field where the questions are complex and the standards for scientific accuracy are high. Autodesk is also named as an early adopter, pointing to engineering design as another vertical moving OpenAI workloads into AWS production.

A Bit of Background

This integration was made possible by a revision to the OpenAI-Microsoft partnership terms in late April 2026. Microsoft had held preferential and exclusive cloud usage rights for OpenAI models through Azure. Once those terms were revised, OpenAI could distribute its latest models through other cloud providers. The integration launched on April 28 in limited preview and reached general availability roughly five weeks later. It is backed by a $50 billion Amazon investment in OpenAI and a multi-year infrastructure deal.

What’s Coming Next

Amazon Bedrock Managed Agents, powered by OpenAI, is listed as coming soon. It will allow teams to deploy production-ready agents built on the OpenAI agent harness, targeting faster execution and reliable steering of long-running tasks. OpenAI’s Daybreak platform, which includes cyber models and Codex Security for secure code review, threat modelling, and patch validation, is also on the roadmap.

What This Means for You

If your team is already building on AWS and has been waiting to bring OpenAI models into production rather than running experiments in a separate environment, the path is now straightforward. The models meet you where your compliance, billing, and tooling already live.

For developers, the Codex integration is particularly worth evaluating. Pay-per-token pricing with no per-seat commitments makes it easy to start small, and the IDE integrations mean it fits into existing workflows rather than requiring a context switch.

The AWS get-started guide has the full technical setup, and the Bedrock OpenAI landing page covers pricing and regional availability in detail.