OpenAI models land on Amazon Bedrock

What AWS and OpenAI announced

Amazon Web Services and OpenAI have put OpenAI’s frontier models on Amazon Bedrock, AWS’s managed AI model service, for the first time. AWS opened limited preview on 28 April 2026, then updated on 1 June to confirm that GPT-5.5, GPT-5.4 and the Codex coding agent are now generally available on the platform. AWS calls it a major expansion of an existing partnership, not a fresh tie-up.

Three things arrive together: OpenAI’s frontier models on Bedrock, the Codex coding agent on Bedrock, and a new managed-agents runtime, currently in limited preview.

What’s actually available today

For teams already on AWS, the access path mirrors any other Bedrock model: AWS identity management, private network connectivity, content guardrails, encryption, and audit logging. Pricing matches OpenAI’s first-party rates with no AWS markup, and usage counts toward existing AWS cloud commitments — the Enterprise Discount Programme pot many UK firms already have in place.

The three pieces, in plain terms:

• OpenAI’s frontier models — GPT-5.5 and GPT-5.4, accessed through the same Bedrock service that already hosts Anthropic, Meta, Mistral and Amazon’s own models.
• Codex on Bedrock — OpenAI’s coding agent, available through the Codex CLI, the desktop app, the VS Code extension, and the Bedrock API. Authentication runs on AWS credentials and inference stays on Bedrock. (Codex’s persistence story sits separately.)
• Bedrock Managed Agents — a new way to deploy OpenAI-powered agents on AWS. In limited preview, with per-agent identity, full action logging, and AWS’s agent compute layer underneath.

AWS points to Box, the content management company, as an early customer. “Enterprises are currently looking to deploy agents to deliver solutions that take their organization into the next phase of AI,” said Ben Kus, Box’s CTO, in the announcement.

4M+people use Codex every week, per the AWS announcement

Why the procurement angle matters in the UK

For most UK small and mid-sized firms this is not a new tool to switch to. The ChatGPT and OpenAI API experience most readers use today still exists, still works, and is still the default for anyone without an AWS relationship. What changes is the buyer conversation.

UK teams already running workloads on AWS — many of which moved data onto the platform in part for data residency and audit reasons — now have a way to put OpenAI’s models on the same bill, under the same controls, against the same committed spend. For regulated buyers, that consolidates procurement rather than adding a new vendor to assess. For finance teams, it lets OpenAI usage compete for the same budget envelope as EC2 or S3, instead of sitting in a separate line item.

It also lowers the cost of switching if OpenAI is the right model for a workload but the procurement path was what blocked adoption. That is the flip side of OpenAI’s broader push this year: an expanded free Academy and certifications at one end, and AWS distribution at the other.

What this means and what to watch

This is a landscape move, not an adopt-it-this-afternoon release for most readers. The threshold questions for a UK firm are simple: are you already an AWS shop with committed spend, and is OpenAI’s frontier capability the thing you wanted but couldn’t route through procurement? If yes, the door is now open and the conversation belongs in your next cloud review. If you are paying OpenAI direct today and your AWS footprint is small, nothing in the announcement changes your workflow. It just gives you a route if the security and procurement picture tightens later.

Two things to watch. First, the managed-agents alignment is the strategically more interesting development: AWS and OpenAI building a shared agent runtime is a notable move, and what surfaces in products over the next two quarters will say a lot about where Bedrock sits relative to OpenAI’s own agent platform. Second, the UK compute story is the parallel track for firms whose data residency points away from the US hyperscalers.

For most UK small teams, this is a story to track, not to act on this month. The right moves remain what they were: pick the model that solves your problem, on the platform that meets your data and procurement needs. Bedrock is now one more platform to keep in the running when those conversations come up. The question worth watching is whether AWS, OpenAI and the broader agent-runtime field converge on a shared production model over the next two quarters.