OpenAI’s Code Agent Just Moved Into Amazon’s House

Five million people use Codex every week. Until this month, every one of those sessions ultimately flowed through infrastructure tied to Microsoft. On June 1, that changed: OpenAI’s frontier models GPT-5.5 and GPT-5.4, along with the Codex coding agent, became generally available on Amazon Bedrock — in both AWS Commercial and GovCloud regions. It is the clearest signal yet that the era of single-cloud frontier AI is over, and it quietly rewrites the procurement math for every enterprise that already runs on AWS.

What Actually Shipped on June 1

Frontier models at first-party prices

One month after the expanded partnership was announced in limited preview, GPT-5.5 and GPT-5.4 are now live in the Amazon Bedrock model catalog, callable through the Responses API. The detail that matters most for buyers: AWS confirmed the per-token rate is identical to buying direct from OpenAI, with no platform markup. GPT-5.5 — OpenAI’s most capable model — is positioned for multi-step agentic work: writing and debugging code across large codebases, analyzing data, generating documents, and operating software across multiple tools until a task is complete.

Codex without the seat license

Codex on Bedrock is the more disruptive half of the release. The agent is available through the Codex App, the Codex CLI, and IDE integrations with Visual Studio Code, JetBrains, and Xcode — but all model inference routes through Amazon Bedrock, staying within the customer’s selected AWS Region for data residency. Pricing is pure pay-per-token: no seat licenses, no per-developer commitments, and usage counts toward existing AWS spend commitments. For engineering leaders who have been blocked by per-seat procurement cycles, that is a meaningful unlock.

The Multi-Cloud Chess Game Behind the Release

Life after Azure exclusivity

This launch is the first major fruit of OpenAI’s post-exclusivity strategy. OpenAI’s exclusive cloud distribution arrangement with Microsoft ended on April 27, 2026, with Microsoft retaining a royalty-free, non-exclusive license to OpenAI’s models through 2032. Barely five weeks later, OpenAI’s flagship models and its fastest-growing product are generally available on Microsoft’s largest competitor. Meanwhile, Microsoft hedged in the other direction at Build, unveiling MAI-Code-1-Flash and a slate of homegrown models to reduce its own reliance on OpenAI.

Why AWS wanted this badly

For AWS, the gap in its catalog was obvious: Bedrock offered Anthropic, Meta, Mistral, and Amazon’s own Nova models, but the most recognizable name in AI was missing. Now AWS can tell its millions of customers they can run OpenAI workloads behind the IAM permissions, VPC and PrivateLink isolation, KMS encryption, and CloudTrail audit logging they already operate. AWS also stressed that prompts and responses are not used to train models and are not shared with model providers — including OpenAI itself. Early enterprise interest spans regulated industries: Amgen’s CTO Sean Bruich called AWS “an important new path” to scale frontier models inside existing governance frameworks, and Autodesk is evaluating Codex for design workflows.

What Comes Next: Managed Agents and Daybreak

Production agents with identities and audit logs

AWS says Amazon Bedrock Managed Agents, powered by OpenAI’s agent harness, is coming soon. Each agent will operate with its own identity, log every action for auditability, and run all inference on Bedrock. Bedrock’s inference engine already captures the full state of each request durably and continuously — if hardware fails mid-call, the request resumes where it left off rather than restarting. That kind of plumbing matters far more for long-running agents than for chatbots.

Security is the next battleground

Both companies flagged Daybreak — OpenAI’s vision for how software is built and defended, which includes cyber models and Codex Security — as a future Bedrock arrival. The pitch: secure code review, threat modeling, patch validation, dependency risk analysis, and remediation guidance inside the everyday development loop, adopted through the security and procurement frameworks security teams already use. With Anthropic expanding Project Glasswing to roughly 150 organizations the same week, frontier labs are now openly competing to own the AI-security layer of the software stack.

Related

• Your AI Agent Just Got Its Own Credit Card
• The AI Lab That Refused Money Just Took $7.4 Billion
• Software Giants Are Quietly Buying the One Thing AI Lacks
• Nvidia Just Walked Into the One Market It Didn’t Own

Sources

1. OpenAI — OpenAI frontier models and Codex are now available on AWS
2. AWS Machine Learning Blog — OpenAI models and Codex on Amazon Bedrock are now generally available
3. The New Stack — The OpenAI-Microsoft reset, decoded: Why AWS may come out ahead