Codex & GPT-5.5 Now Generally Available on Amazon Bedrock

Codex and OpenAI Frontier Models Reach General Availability on Amazon Bedrock

After a limited preview launched in late April 2026, OpenAI's flagship models and Codex have officially reached general availability on Amazon Bedrock as of June 1, 2026. Enterprise teams and developers can now access Codex at production scale without waitlists, using the full suite of AWS security, governance, and billing controls they already rely on.

What's Available

GPT-5.5 is OpenAI's most capable frontier model, built for multi-step agentic tasks including code debugging, complex data analysis, and document generation. It operates on Bedrock's next-generation inference engine, combining high throughput with the reliability and compliance guarantees enterprises expect from AWS.

GPT-5.4 is also available via the Responses API, providing a strong balance of capability and cost for organizations that don't require the full power of GPT-5.5.

Codex, OpenAI's AI-powered coding agent used by over 5 million developers weekly, now routes all inference through Amazon Bedrock. Codex handles writing, refactoring, debugging, and testing across large codebases, maintaining deep repository context throughout long-running tasks. It integrates with the Codex CLI, the Codex desktop app, and IDE extensions for Visual Studio Code, JetBrains, and Xcode.

How It Differs from the April Limited Preview

The April 28, 2026 limited preview required signup and restricted access to a subset of AWS customers. The June 1 GA removes those constraints entirely, making Codex and GPT-5.5/GPT-5.4 available to all AWS customers at full production scale with committed SLAs and without a waitlist.

AWS-Native Infrastructure

Running Codex through Bedrock provides enterprise-grade controls that integrate directly with the AWS ecosystem:

• IAM authentication: Teams authenticate using existing AWS IAM roles and policies, eliminating separate OpenAI credential management.
• VPC isolation: All inference traffic stays within a customer's own AWS VPC.
• KMS encryption: Prompts, code, and responses are encrypted using AWS Key Management Service.
• CloudTrail logging: Every Codex interaction is logged through CloudTrail for compliance and audit purposes.
• Data privacy: User prompts and model responses are not used for OpenAI model training.

Pricing and Billing

Codex on Bedrock operates on pay-per-token pricing with no seat licenses and no per-developer commitments. Pricing matches OpenAI's direct rates exactly, with no additional AWS fees. Usage counts toward existing AWS spending commitments, meaning organizations with Enterprise Discount Programs or AWS Marketplace agreements can apply those commitments to Codex inference.

Looking Ahead

AWS and OpenAI have signaled that additional capabilities are in development for the Bedrock integration, including Bedrock Managed Agents powered by OpenAI and Daybreak, OpenAI's security-focused vision tooling. The GA marks the start of an expanded multi-cloud strategy for OpenAI, giving enterprise teams infrastructure choice without model lock-in.