- One of OpenAI’s most advanced frontier models, GPT-5.5, is now officially available on Amazon Bedrock, with pricing consistent with OpenAI’s official rates.
- Codex is now officially available on Amazon Bedrock, adopting a per-token billing model. Inference runs are processed through Amazon Bedrock, and usage counts toward customers’ existing AWS committed usage.
BeijingJune 2, 2026 /PRNewswire/ — Amazon Web Services (AWS) announced that, one month after deepening its partnership with OpenAI, GPT-5.5, GPT-5.4, and Codex are now officially available on Amazon Bedrock. Enterprises can directly access frontier large language models or leverage OpenAI’s coding agents to advance software development. Amazon Bedrock is a service for building and running AI applications and agents at production scale. OpenAI models on Amazon Bedrock run on Amazon Bedrock’s next-generation inference engine, delivering high performance, high reliability, and high security.
The most powerful OpenAI models currently available on Amazon Bedrock
GPT-5.5 understands intent faster and can autonomously complete multi-step tasks. From writing and debugging large codebases, to data analysis, document and spreadsheet generation, and cross-tool software operations, it executes tasks from start to finish. The most notable improvements lie in intelligent coding and knowledge work—scenarios that require maintaining context over extended periods and persistently advancing tasks. This is precisely where GPT-5.5 achieves its breakthrough.
GPT-5.5 and GPT-5.4 are both designed for complex, multi-step tasks. They are now available in the Amazon Bedrock model catalog and can be invoked via the Responses API. Billing is exactly the same as using OpenAI directly, with no additional fees.
At the infrastructure level, the Bedrock inference engine provides each user with an independent, isolated queue, automatically managing capacity to ensure stable performance even under high loads. During each request’s execution, the complete state is continuously captured and persisted: in the event of hardware failure or node restart, the request resumes directly from where it was interrupted, without needing to start over. In terms of security and compliance, each invocation fully inherits AWS’s existing governance and control framework: Amazon IAM permission management, Amazon VPC and PrivateLink network isolation, Amazon KMS encryption, and Amazon CloudTrail full-chain audit logs. Prompt and response content is neither used for model training nor shared with the model provider. GPT-5.5 and GPT-5.4 on Amazon Bedrock are subject to all of the above protective measures.
Sean Bruich, Senior Vice President and Chief Technology Officer at Amgen, said: “At Amgen, we are committed to responsibly applying advanced AI to accelerate the development of potential new therapies, while equipping our teams with cutting-edge tools. The problems in life sciences are complex, with extremely high demands on scientific accuracy and decision quality. The improvements in capability, quality, and consistency offered by OpenAI’s GPT-5.5 and frontier models are exactly what this field needs. Deploying these models on AWS allows Amgen to explore and scale these capabilities within an enterprise-grade responsible AI framework—covering safety, governance, and operations—which represents an important new pathway.”
Accelerating software development with Codex on Amazon Bedrock
Codex is OpenAI’s coding agent for AI-driven software development. Over 5 million users per week use Codex for code writing, refactoring, debugging, testing, and validation across large codebases. Codex maintains contextual coherence across entire code repositories, performs deep reasoning on ambiguous error messages, proactively validates hypotheses through tools, and, based on an understanding of the overall system structure, applies changes to all relevant surrounding code. When switching to GPT-5.5 as the inference backbone, the efficiency and quality of completing equivalent tasks have significantly improved compared to previous versions.
Codex on Amazon Bedrock supports direct use via the Codex App, Codex CLI, and plugins for mainstream IDEs such as Visual Studio Code, JetBrains, and Xcode. All model inference requests are uniformly processed through Amazon Bedrock. Inference computing is completed within designated regions, meeting data residency compliance requirements. It is billed per token, with no need to purchase seat licenses or pre-commit to usage based on developer headcount, making it easy to get started and flexible to scale.
Ritesh Bansal, Vice President of Analytics Data, Agentic AI, and AI/ML Platform at Autodesk, said: “Autodesk is a technology platform serving designers and builders worldwide. Workflows like architectural design involve frequent iterations, requiring precise collaboration and continuous refinement across teams. With OpenAI models and Codex now available on Amazon Bedrock, Autodesk’s teams are evaluating how to leverage frontier AI capabilities and AI-driven development tools on AWS’s secure, scalable infrastructure to accelerate development processes and provide better decision support for our customers.”
Future plans
In the previous announcement of the deepened partnership, AWS announced the upcoming launch of Amazon Bedrock Managed Agents powered by OpenAI, allowing users to deploy production-ready agents built on the OpenAI Agent framework, achieving faster execution speeds, sharper reasoning capabilities, and reliable scheduling for long-running tasks. Each agent will have an independent identity, recording every step to ensure auditability, and running all model inference on Amazon Bedrock.
AWS will continue to extend OpenAI’s latest capabilities to Amazon Bedrock. This includes Daybreak—OpenAI’s vision for transforming how software is built and defended. Daybreak includes a cybersecurity model and Codex Security, designed to help cybersecurity defenders identify vulnerabilities, review code risks, and provide remediation guidance throughout the entire development lifecycle. When Daybreak becomes available on Amazon Bedrock, security teams will be able to adopt this capability through the governance and operational frameworks they already use on AWS.