OpenAI Brings GPT-5.5 and Codex to Amazon Bedrock — One Day After Microsoft Exclusivity Ends (April 28, 2026)

OpenAI on April 28, 2026 announced that its frontier models, the Codex coding harness and a new managed-agents service are available in limited preview on Amazon Bedrock — the first time GPT-class models from OpenAI have been served from a cloud other than Microsoft Azure. The launch arrived less than 24 hours after Microsoft and OpenAI restructured their partnership to end Azure's API-services exclusivity and clear the legal path for OpenAI's reported $38–$50 billion compute deal with AWS.

What Happened

On April 27, Microsoft and OpenAI quietly rewrote the terms of their multi-year alliance: Azure exclusivity for OpenAI's API products is gone, Microsoft no longer gets a revenue share on OpenAI's third-party cloud usage, and Microsoft drops its standing first-refusal right to host every new OpenAI workload. The next morning AWS published an official announcement on its newsroom and OpenAI co-published a post on openai.com confirming that three things are now live in limited preview on Amazon Bedrock:

Key Details

• OpenAI models on Bedrock: GPT-5.5 and GPT-5.4 are the launch models, served behind the standard Bedrock InvokeModel and Converse APIs. AWS PrivateLink, IAM, CloudTrail, encryption-at-rest and Bedrock Guardrails all apply unmodified.
• Codex on AWS: Customers can point the Codex CLI, the Codex desktop app and the Codex VS Code extension at a Bedrock endpoint via AWS_BEARER_TOKEN_BEDROCK, paying through their AWS bill rather than an OpenAI API key.
• Bedrock Managed Agents: A new agent runtime, billed separately, that handles tool use, orchestration and governance for OpenAI-powered agents. AWS describes it as the "harder parts of deployment" productised, including built-in checkpoints, tracing and IAM-scoped tool execution.
• Limited preview gating: Access is request-only and tied to existing AWS commits; AWS did not publish per-token pricing at launch.
• Microsoft exclusivity ended: Microsoft retains its 27% economic stake in OpenAI Group PBC and a first-refusal on stateful services like ChatGPT, but lost the API-services exclusivity it had held since 2023.

What Developers and Users Are Saying

The Hacker News thread on the launch crossed the front page within an hour, with the most upvoted comments focused on the practical implications for multi-cloud LLM routing. Several developer-tooling founders said they were already updating their SDKs to treat Bedrock as a first-class OpenAI provider alongside Azure and the OpenAI API. On the AWS subreddit, reaction skewed positive: enterprise developers flagged that this is the cleanest path yet to running GPT-5.5 inside a VPC under existing IAM controls, without negotiating a separate Azure tenant. Critics pointed out that limited-preview access without published pricing means evaluation is still gated, and that the Codex CLI integration relies on a token environment variable rather than a deeper SDK hand-off — a friction point that will likely be smoothed out before general availability.

What This Means for Developers

For the first time, OpenAI's frontier models can be procured, billed and operated through an AWS contract. The immediate practical effects: enterprise architects who already standardised on Bedrock can drop GPT-5.5 into existing pipelines without onboarding Azure; agent platforms can use Bedrock Managed Agents to ship governance-ready OpenAI agents without rebuilding their orchestration layer; and Codex users at AWS-heavy shops can pay through their existing committed-use discount. SDK maintainers should expect a wave of pull requests adding Bedrock as a routing target, and SREs should plan for the new aws.bedrock.openai.* CloudWatch metric namespace that AWS introduced alongside the launch.

What's Next

AWS has not committed to a general-availability date for OpenAI on Bedrock, but the official posts say the limited preview will expand in "the coming weeks" and that additional OpenAI models will be added over time. OpenAI's reported $38–$50 billion AWS compute commitment runs over seven years and includes provisioning of hundreds of thousands of Nvidia GPUs across multiple AWS regions. Watch for: published Bedrock pricing for GPT-5.5, the first GA models, the addition of OpenAI's open-weight gpt-oss-120b and gpt-oss-20b models on Bedrock (which AWS confirmed are coming), and how Microsoft responds — Azure historically held a 6–12 month head-start on every new OpenAI model and will need to defend that position.

Sources

• OpenAI — OpenAI models, Codex, and Managed Agents come to AWS (primary)
• Amazon — AWS and OpenAI announce expanded partnership (primary)
• AWS What's New — Limited preview launch notes
• CNBC — OpenAI brings models to AWS after ending exclusivity with Microsoft
• GeekWire — One day after Microsoft exclusivity ends
• The Register — OpenAI jumps into Amazon's Bedrock
• Hacker News — Discussion thread (front page)