OpenAI has made its frontier models and Codex generally available on Amazon Web Services, converting an April limited preview into a production-ready launch for enterprise customers building generative AI applications on Amazon Bedrock.

The June 1 announcement gives AWS customers access to OpenAI’s GPT-5.5 and GPT-5.4 models, along with Codex, OpenAI’s software engineering agent, through AWS infrastructure and operating controls. OpenAI positioned the launch as a way for enterprises to bring advanced AI into production without having to create new procurement, security, billing, governance or deployment processes outside the cloud systems they already use.

The development is a notable expansion in the commercial reach of OpenAI’s models. For years, Microsoft Azure was the central enterprise cloud distribution path for OpenAI technology. Microsoft said in April that its license to OpenAI intellectual property would become non-exclusive, while Microsoft would remain OpenAI’s primary cloud partner and continue to have licensing rights for OpenAI models and products through 2032. That change allowed OpenAI to serve products to customers across other cloud providers, including AWS.

For AWS, the launch strengthens Amazon Bedrock’s role as a managed platform for accessing and deploying multiple foundation models under a common enterprise control layer. AWS has marketed Bedrock as a way for organizations to choose among model providers while retaining consistent security, governance, orchestration and cost-management practices. With OpenAI now generally available, Bedrock customers can evaluate and deploy OpenAI models alongside models from other providers already present in the AWS ecosystem.

AWS said GPT-5.5, GPT-5.4 and Codex are now generally available on Amazon Bedrock, with pricing that matches OpenAI’s first-party rates and usage that counts toward AWS commitments. The company said GPT-5.5 is designed for demanding workloads including agentic coding, data analysis and multi-step autonomous tasks, while GPT-5.4 is positioned for customers seeking a balance of capability and price performance.

The general availability milestone matters because many enterprises have moved beyond proof-of-concept work but remain constrained by compliance reviews, data governance requirements, identity controls and cost allocation systems. Deploying AI models through an existing AWS account can reduce friction for companies that already standardize infrastructure procurement and risk management around AWS. In practice, that may accelerate adoption in sectors where AI experimentation is common but production deployment requires stricter internal approvals.

OpenAI said customers can now bring its capabilities into AWS environments using the controls their teams already trust. The company said the offerings are available in both commercial and GovCloud regions, a detail that signals OpenAI and AWS are targeting regulated and public-sector-adjacent workloads as well as mainstream corporate software development.

The technical route is Amazon Bedrock, AWS’s managed generative AI service. AWS said the OpenAI models run on Bedrock’s next-generation inference engine, which it described as built for performance, reliability and security. Customers can call the models through the Responses API on Bedrock, giving developers a pathway that is familiar to OpenAI users while integrating with AWS infrastructure.

Codex is the other major part of the launch. OpenAI describes Codex as a software engineering agent used to write, review, debug and modernize code. AWS said Codex can be used through the Codex App, Codex CLI and integrated development environments including Visual Studio Code, JetBrains and Xcode, with inference routed through Amazon Bedrock. That turns Codex into a cloud-governed development tool rather than a separate service operating outside the enterprise’s AWS controls.

The software development angle is commercially important for AWS and OpenAI because coding assistants and agentic development tools are among the clearest enterprise AI use cases. Companies can measure developer productivity, code modernization velocity, test generation, documentation output and migration support more directly than many broad knowledge-work applications. Bringing Codex into AWS gives technology organizations a way to use OpenAI’s coding capabilities while keeping inference, identity and governance aligned with existing cloud policies.

AWS emphasized that Codex inference can be routed through Bedrock with security protections such as AWS Identity and Access Management, VPC isolation and encryption. The company also said processing remains within the selected Bedrock Region, an important requirement for companies with data residency obligations. These details are central to the value proposition because enterprise AI buying decisions increasingly turn on governance and infrastructure assurances, not only benchmark performance.

The launch also changes the competitive posture of AWS in the AI cloud market. Amazon has invested heavily in Bedrock as a multi-model platform, while also building its own AI infrastructure and model offerings. Adding OpenAI’s frontier models gives AWS customers another high-profile model family inside the same purchasing and deployment channel. That could help AWS counter the perception that Microsoft Azure has the most direct OpenAI advantage, while allowing Amazon to keep positioning Bedrock as a broad model marketplace rather than a single-vendor AI stack.

For OpenAI, AWS general availability expands distribution into one of the world’s largest enterprise cloud customer bases. The company has increasingly presented its business products not only as consumer-facing AI tools but as production infrastructure for enterprises. The AWS launch supports that strategy by embedding OpenAI models into the operational environment used by many large companies for databases, analytics, application hosting, security monitoring and software delivery pipelines.

The announcement also underscores the shift from model access as a standalone product to model access as part of an enterprise platform. Businesses adopting generative AI often need logging, permissioning, auditability, deployment controls, regional data handling and cost tracking. By offering OpenAI models inside Bedrock, AWS can bundle model access into the broader cloud operating model. OpenAI, in turn, can reach customers that might otherwise face internal barriers to sending workloads to a separate AI provider account.

The launch builds on an April 28 partnership expansion in which OpenAI and AWS introduced OpenAI models on AWS, Codex on AWS and Amazon Bedrock Managed Agents powered by OpenAI in limited preview. At that time, OpenAI said the partnership would help organizations use OpenAI capabilities across application development, software engineering and agentic workflows while building within AWS infrastructure, security, governance and procurement systems. The June 1 update moves the core model and Codex offerings from preview into general availability.

Amazon Bedrock Managed Agents powered by OpenAI were part of the earlier partnership expansion, although the June 1 general availability announcement focused on OpenAI frontier models and Codex. The broader direction remains clear: AWS and OpenAI are trying to support agentic workflows that can maintain context, use tools, execute multi-step processes and operate inside enterprise controls. That aligns with the wider industry push from chat interfaces toward AI agents embedded in business processes.

Customer examples included in OpenAI’s announcement point to the kinds of industries the companies are targeting. Amgen said access to OpenAI models on AWS gives it another path to explore advanced AI within responsible AI, security, governance and operational frameworks. Autodesk said it is evaluating how frontier AI capabilities and AI-powered development tools on AWS infrastructure can support development workflows and decision-making for its customers. Those examples reflect demand from both regulated life sciences and design-software environments, where accuracy, workflow integration and governance are central concerns.

The pricing structure is another strategic lever. AWS said pricing matches OpenAI first-party rates and that usage counts toward AWS commitments. For large cloud customers, that matters because many negotiate multi-year spending commitments with AWS. If OpenAI model usage can be counted toward those commitments, customers may face fewer budgetary barriers than they would when procuring a separate service outside their cloud agreement.

The model availability is region-specific at launch. AWS said GPT-5.5 is available in the US East (Ohio) Region, while GPT-5.4 is available in US East (Ohio) and US West (Oregon), with customers directed to AWS regional documentation for future updates. Regional availability will be an important factor for multinational enterprises that need low latency, data residency and compliance alignment across geographies.

The launch may also increase pressure on rival cloud platforms and model providers. Enterprise customers are increasingly seeking flexibility to compare models by task, cost, latency, governance profile and vendor risk. If AWS can offer OpenAI alongside other major providers through a consistent control plane, it may become easier for customers to run model selection and orchestration strategies that do not depend on a single provider. That dynamic could intensify competition on inference performance, pricing and operational tooling.

For Microsoft, the broader distribution of OpenAI technology changes but does not eliminate its strategic position. Microsoft remains deeply tied to OpenAI through licensing and product integration, including Azure AI services and Microsoft 365 Copilot. However, OpenAI’s availability on AWS reduces the exclusivity advantage that helped distinguish Azure for companies specifically seeking OpenAI access. The competitive question now shifts toward which cloud provides the best total enterprise environment for deploying OpenAI-powered applications at scale.

The timing also reflects a broader enterprise AI transition in 2026. Companies are no longer evaluating only whether large language models are useful; they are evaluating where those models should live, how they should be governed and how agentic systems should interact with internal software. OpenAI and AWS are presenting the general availability launch as an answer to that infrastructure question: advanced models and coding agents available inside familiar cloud controls.

The next phase will depend on customer adoption and execution. General availability gives enterprises a stronger basis for production deployment, but large-scale AI rollout still depends on internal data architecture, application integration, developer training, risk controls and measurable return on investment. OpenAI and AWS have reduced one major barrier by placing OpenAI capabilities inside AWS workflows. Whether that translates into sustained cloud revenue and broader enterprise AI deployment will depend on how quickly customers turn model access into durable production systems.