OpenAI’s GPT-5.5, GPT-5.4, and the Codex coding agent are now generally available on Amazon Bedrock. This release comes one month after the initial announcement of an expanded partnership between AWS and OpenAI, allowing organizations to integrate frontier models into production-scale applications and agents. Amazon Bedrock provides an inference engine that ensures predictable performance and reliability, even under heavy load, by managing isolated queues and automated capacity. Users can access these models through standard AWS governance frameworks, including IAM permissions, VPC isolation, and AWS CloudTrail auditing. Notably, data used with these models on Bedrock is not utilized for further training and remains private to the user account.

GPT-5.5 serves as the most advanced frontier model available through the platform, designed to manage multi-step, autonomous tasks. It is optimized for agentic coding and knowledge work, which require maintaining context and taking persistent action over time. Both GPT-5.5 and GPT-5.4 are billed at parity with OpenAI’s direct first-party rates, with no additional fees for using the Bedrock infrastructure. Amazon's system captures the full state of requests durably, ensuring that tasks can resume from a point of interruption if a hardware node fails during execution.

Codex is also available as an OpenAI coding agent to support software development across large codebases. It is accessible through the Codex App, the Codex CLI, and IDE integrations including Visual Studio Code, JetBrains, and Xcode. The tool is designed to reason through ambiguous failures and check code assumptions using external tools. Pricing for Codex on Bedrock follows a pay-per-token model without requiring seat licenses or developer commitments. Future updates to the platform will include Amazon Bedrock Managed Agents, which are expected to deliver improved steering of long-running tasks, as well as the integration of Daybreak, a forthcoming security-focused suite for vulnerability identification and code risk remediation.

OpenAI’s GPT-5.5, GPT-5.4, and the Codex coding agent are now generally available on Amazon Bedrock. This release comes one month after the initial announcement of an expanded partnership between AWS and OpenAI, allowing organizations to integrate frontier models into production-scale applications and agents. Amazon Bedrock provides an inference engine that ensures predictable performance and reliability, even under heavy load, by managing isolated queues and automated capacity. Users can access these models through standard AWS governance frameworks, including IAM permissions, VPC isolation, and AWS CloudTrail auditing. Notably, data used with these models on Bedrock is not utilized for further training and remains private to the user account.

GPT-5.5 serves as the most advanced frontier model available through the platform, designed to manage multi-step, autonomous tasks. It is optimized for agentic coding and knowledge work, which require maintaining context and taking persistent action over time. Both GPT-5.5 and GPT-5.4 are billed at parity with OpenAI’s direct first-party rates, with no additional fees for using the Bedrock infrastructure. Amazon's system captures the full state of requests durably, ensuring that tasks can resume from a point of interruption if a hardware node fails during execution.

Codex is also available as an OpenAI coding agent to support software development across large codebases. It is accessible through the Codex App, the Codex CLI, and IDE integrations including Visual Studio Code, JetBrains, and Xcode. The tool is designed to reason through ambiguous failures and check code assumptions using external tools. Pricing for Codex on Bedrock follows a pay-per-token model without requiring seat licenses or developer commitments. Future updates to the platform will include Amazon Bedrock Managed Agents, which are expected to deliver improved steering of long-running tasks, as well as the integration of Daybreak, a forthcoming security-focused suite for vulnerability identification and code risk remediation.