AWS posted on Monday that OpenAI’s GPT-5.5, GPT-5.4, and Codex are now generally available on Amazon Bedrock. The models run through the new Bedrock Responses API at the same per-token rates OpenAI charges through its own endpoint. No platform markup, no margin stack. GPT-5.5 ships in US East (Ohio), GPT-5.4 ships in Ohio and US West (Oregon), and Codex is the coding-agent product OpenAI has been quietly turning into a first-class billable line.
The interesting thing is not that GPT-5.5 is on Bedrock. It is that Anthropic’s Mythos preview shipped on the same platform four days ago, and AWS just put both frontier labs on the same procurement surface with zero per-token margin. AWS does not need the margin. AWS gets paid on the inference compute, the storage, the egress, the surrounding IAM and KMS and CloudTrail line items, and the fact that customers who run their AI workloads on Bedrock are also running their databases, their queues, and their identity on AWS. The frontier model is a loss leader for the rest of the bill.
The Codex line is the one to underline. OpenAI’s coding agent was previously a direct-only product, which meant enterprise procurement teams who had spent six months wiring up Bedrock-based controls for chat models had to spin up a parallel approval process for the agent that writes code into their repos. As of Monday, Codex inherits the entire Bedrock security envelope: same IAM, same logging, same data-residency story, same single-tenant guarantees. The procurement unlock is doing the work the marketing copy isn’t.
The structural read is that the hyperscaler distribution game has crystallized faster than the model-vendor differentiation game. Azure has OpenAI deeply integrated. Google Cloud has Gemini natively. AWS is now positioned as the neutral one, which is the position that historically wins B2B once the dust settles. The frontier labs are racing to ship better models. The cloud platforms are racing to make it not matter which one you pick.