AWS makes OpenAI models and Codex a native Bedrock option

AWS made GPT-5.5, GPT-5.4 and Codex generally available through Amazon Bedrock on 1 June 2026. The launch gives organisations another way to use OpenAI systems without moving identity, procurement, billing and operational controls away from their existing AWS setup. That makes this less a simple model listing and more a contest over who owns the enterprise control room.

AWS AI Watch analysis

What happened

AWS says GPT-5.5 and GPT-5.4 can be used in production workloads through Bedrock's Responses API. Codex can route its model inference through Bedrock from the Codex App, command-line tool and integrations for Visual Studio Code, JetBrains and Xcode. AWS says model access inherits controls including IAM permissions, private networking, encryption and audit logging.

The commercial pitch is equally deliberate. AWS says model pricing matches OpenAI's first-party per-token rates, while Codex usage can count toward existing AWS commitments. OpenAI's own announcement frames the same arrangement around reducing friction in security review, compliance, procurement and deployment. Both accounts are first-party descriptions, not independent measurements of cost, reliability or deployment speed.

Why it matters

Enterprises choosing a model are also choosing where permissions, logs, regional deployment and spending controls live. Bedrock lets AWS keep that surrounding relationship even when the model comes from a partner. For OpenAI, the bargain broadens access to customers whose buying and governance machinery already runs through AWS. For buyers, it creates a credible additional route, but not automatic feature parity with OpenAI's direct service.

Our read

The clever bit is not putting another model on a menu. It is letting the security team keep the familiar key cabinet while developers try a different engine. The shed question is whether that convenience survives the first awkward requirement that one route supports and the other does not.

What to watch

• Real workload comparisons covering features, regional availability, latency and total cost.
• Whether AWS commitments materially change purchasing decisions for Codex and OpenAI models.
• How quickly future OpenAI capabilities reach Bedrock after appearing through direct channels.
• Whether enterprises standardise on a cloud control plane or preserve multiple deployment routes.

Discussion spark: Does the durable enterprise advantage belong to the model maker, or to the cloud provider controlling identity, data, routing and procurement?

not affiliated with or endorsed by Amazon Web Services (AWS)