Amazon Web Services has made OpenAI’s GPT-5.5, GPT-5.4, and Codex models, along with new OpenAI-powered Bedrock Managed Agents, generally available in Amazon Bedrock, the company announced today. The move follows a limited preview that began in spring 2026 and marks the first time these advanced models are offered outside of OpenAI’s own platform, according to a statement from AWS.

The New AI Toolkit on AWS Bedrock

The general availability release brings four key additions to the Bedrock service: GPT-5.5, GPT-5.4, Codex, and Bedrock Managed Agents. GPT-5.5 is OpenAI’s most advanced general-purpose language model, offering improved reasoning, instruction following, and multilingual capabilities over earlier versions. GPT-5.4 is a slightly earlier iteration that balances performance with cost efficiency, giving enterprises more budget flexibility. Codex, the model fine-tuned for programming tasks, can generate, explain, and debug code across dozens of languages, making it a powerful tool for developer productivity. The new Managed Agents feature lets organizations build and orchestrate complex AI workflows without managing the underlying infrastructure—agents can autonomously perform multi-step tasks, call APIs, and interact with enterprise data sources, all governed by Bedrock’s security and monitoring framework.

These models are now accessible via the Bedrock API, AWS CLI, SDKs, and the AWS Management Console, just like the existing Titan, Claude, and third-party models already hosted on the service. AWS has integrated the models with its broader suite, including Amazon CloudWatch for observability and AWS Identity and Access Management (IAM) for fine-grained access control, ensuring that enterprises can safely deploy generative AI at scale.

Why This Matters for Windows-Centric Enterprises

Windows powers the vast majority of enterprise desktops, and its development ecosystem—Visual Studio, VS Code, .NET, PowerShell—is deeply intertwined with cloud backends. For organizations that have standardized on Windows but use AWS as their primary cloud provider, this release fills a critical gap. Previously, accessing OpenAI’s latest models typically meant using the Azure OpenAI Service, which, while deeply integrated with Microsoft’s stack, forced a multi-cloud strategy that many IT leaders wanted to avoid. Now, Windows shops can keep all their AI workloads within AWS, simplifying network architecture, billing, and compliance.

For developers building Windows-native or cross-platform applications, Codex on Bedrock can transform coding workflows. A .NET developer, for example, can integrate Codex into a custom Visual Studio extension to get context-aware code completions, automatically generate unit tests, or refactor legacy codebases. IT administrators managing Windows Server environments can use Bedrock Managed Agents to automate incident response: an agent could monitor logs, diagnose a failing service, open a ticket in ServiceNow, and even apply a tested fix—all within the same cloud fabric that hosts the organization’s Active Directory or SQL Server instances.

Corporate governance is another big win for Windows enterprises. Many run Windows 11 with strict compliance requirements, such as HIPAA or FedRAMP. Bedrock’s model access is metered through the same IAM roles that control access to S3 buckets or EC2 instances, and CloudWatch can log every model invocation for audit trails. That coherence is especially appealing to security teams that already rely on AWS’s security tools for their Windows workloads.

The Road to Multi-Cloud OpenAI

The partnership between Microsoft and OpenAI has long been the cornerstone of enterprise AI—Microsoft’s multibillion-dollar investment gave Azure exclusive cloud provider rights for reselling OpenAI’s models. That arrangement led to the Azure OpenAI Service, which became the default for enterprises wanting GPT-4 and later models. AWS, meanwhile, leaned heavily on its own Titan models and a close partnership with Anthropic, the maker of Claude. For a time, Bedrock appeared to be a curated marketplace for non-OpenAI models, while Azure was the only cloud where you could officially run OpenAI’s latest.

That started to change in early 2026, when reports surfaced that OpenAI was negotiating with AWS to expand its distribution. By spring 2026, AWS announced the limited preview of GPT-5.5, GPT-5.4, and Codex on Bedrock, signaling a significant shift. Analysts saw it as OpenAI hedging against over-reliance on Microsoft, while AWS saw an opportunity to blunt Azure’s AI advantage. The general availability, effective today, confirms that OpenAI is now a multi-cloud vendor, and AWS has caught up—at least in terms of model freshness.

Notably, the preview period was unusually short for enterprise cloud services, suggesting strong customer demand. Early adopters in the preview included global financial institutions and automotive manufacturers, according to AWS’s earlier statements, which used GPT-5.5 for document summarization and contract analysis while keeping data within their existing AWS regions.

How to Get Started with the New Models on Bedrock

If your organization already has an AWS footprint, enabling these models requires minimal new setup. Administrators should first review the model-specific pricing pages on AWS’s website, as costs vary by input and output token volume; GPT-5.5, for example, may command a premium over older models. Access must be explicitly enabled via the Bedrock console or API, and IAM policies need to be updated to allow the bedrock:InvokeModel and related actions for the new model IDs.

Developers can test the models immediately using the Bedrock Playground, a no-code interface in the AWS Console. From there, you can send prompts to GPT-5.5 or Codex and compare outputs with other models housed in Bedrock, such as Claude or Llama. For production integration, AWS provides SDKs for Python, JavaScript, Java, and .NET—making it straightforward to build a C# application that calls Codex to analyze code quality or generate documentation.

Managed Agents involve a bit more planning. You’ll need to define an agent’s “action groups”—the functions it can call—and link it to a knowledge base for retrieval-augmented generation (RAG). AWS provides a visual builder in the console, but teams can also define agents programmatically using CloudFormation or the AWS CDK, which is especially useful for Windows DevOps teams that favor infrastructure-as-code.

Before going live, review the shared responsibility model: AWS secures the infrastructure, but you’re responsible for prompt engineering, data handling, and output validation. Bedrock does not use customer data to train base models, addressing a top enterprise concern. For additional guardrails, you can apply Bedrock’s content filtering and redaction for personally identifiable information (PII), which is critical for regulated industries.

The Competitive Landscape Ahead

Today’s announcement reshuffles the enterprise AI deck. Microsoft’s Azure OpenAI Service now has a direct competitor on AWS, and enterprises can play the two off each other for better pricing or service-level agreements. We can expect Azure to emphasize its unique integration with Microsoft 365 Copilot and the Windows ecosystem as differentiators, while AWS will tout Bedrock’s model diversity and unified management plane.

For Windows users, the choice may come down to tooling. If your development lifecycle is deeply embedded in Azure DevOps, GitHub, and Visual Studio with IntelliCode, Azure retains a frictionless advantage. But if you’re running a mixed Windows/Linux environment on AWS or need the bedrock of Bedrock’s governance features, the new OpenAI models remove a major barrier. Expect rapid iteration: as GPT-5.5 and Codex mature, AWS may introduce fine-tuning options, model caching for low-latency scenarios, and deeper integrations with services like Amazon Kendra for enterprise search. The one certainty is that enterprise AI is no longer a single-cloud story.