What changed
AWS has announced that developers can now use OpenAI Codex with Amazon Bedrock, enabling them to explore unfamiliar code, make changes, and handle repetitive engineering tasks. The integration allows Codex to use Amazon Bedrock for inference with supported OpenAI models, fitting into existing AWS identity, IAM, regional, and observability practices.
What this means in practice
The integration begins with a small, least-privilege pilot, allowing developers to use their existing workforce identity provider for sign-in. AWS Identity and Access Management (IAM) Identity Center provisions users and groups, issuing temporary AWS credentials used by Codex. The examples use GPT-5.6 Sol, and developers are advised to check if Sol, Terra, or Luna is available in the chosen AWS Region and inference path.
The recommended default is to start with Native AWS Access for a small pilot. Developers can choose between AgentCore Gateway for AWS-managed gateway infrastructure or an LLM Gateway for enforcing controls in the request path. A custom integration is an advanced option for those looking to build their own gateway.
Limitations and availability
Codex still runs on the developer workstation, and the main design choice is how its model requests reach Amazon Bedrock. The examples come from the open source Guidance for Codex on AWS repository, which includes quick starts for various access paths. The integration requires developers to review Supported models by AWS Region in Amazon Bedrock to confirm model availability.
The supplied announcement does not specify exact performance benchmarks or independent testing results. The integration is currently in a pilot phase, and developers are advised to start with a small, least-privilege setup before expanding access.
The rollout path depends on the organization's needs, with Native AWS Access being the usual starting point when AWS identity and IAM controls are sufficient for the pilot.
