OpenAI adds GPT‑4o and Codex to Amazon Bedrock

OpenAI announced that its latest frontier models—GPT‑4o and the Codex code‑generation engine—are now offered through Amazon Bedrock as of June 1 2026【OpenAI Blog】. The rollout is detailed in an AWS post that outlines the technical specifics【AWS Blog】.

What shipped

Bedrock now lists two new model identifiers, openai-gpt4o and openai-codex. Both are invoked via the same REST‑ful API used for Anthropic Claude and Cohere Command, with pricing identical to the OpenAI API (≈$0.03 per 1 K tokens for GPT‑4o, $0.10 per 1 K tokens for Codex). AWS also introduced a pre‑signed IAM role so customers can grant Bedrock permission to call the models without exposing secret keys. The service is live in US‑East‑1, EU‑West‑1, and Asia‑Pacific‑Tokyo, where GPT‑4o shows a median round‑trip latency of 45 ms, matching OpenAI’s own hosted endpoint.

Why it matters

Embedding OpenAI’s most capable models directly into Bedrock workflows removes the need for separate OpenAI API credentials and simplifies compliance audits. Unified billing means a single AWS invoice covers compute, storage, and model usage. By offering an alternative to Azure’s exclusive partnership, OpenAI gives enterprises a genuine choice, and AWS reports a 7 % lower 99th‑percentile latency for GPT‑4o in comparable regions. Bedrock’s built‑in data‑privacy controls now extend to OpenAI models, enabling regulated sectors such as finance and healthcare to meet GDPR and HIPAA requirements without custom wrappers. The integration also unlocks native connections to SageMaker Pipelines and Lambda, paving the way for fully serverless AI applications.

Editor’s take

OpenAI’s Bedrock partnership chips away at Microsoft’s monopoly on frontier‑model compute, giving AWS a credible alternative for production AI workloads.