OpenAI GPT-5.5 and Codex on AWS: 5 Facts About the Bedrock GA Launch

Amazon and OpenAI just made a quiet but massive move. OpenAI’s frontier models — GPT-5.5, GPT-5.4, and the Codex coding agent — are now generally available on Amazon Bedrock. That means 5 million Codex users can now run their workloads on AWS infrastructure.

TL;DR: OpenAI’s GPT-5.5, GPT-5.4, and Codex coding agent are now generally available on Amazon Bedrock. Codex brings a 5-million-user base to AWS infrastructure, giving enterprises access through existing security and compliance workflows. The GA launch lets teams deploy frontier models in production on Bedrock’s high-performance inference engine.

What Models Did OpenAI Bring to Amazon Bedrock?

OpenAI made three distinct offerings generally available on Amazon Bedrock as of the GA launch: GPT-5.5, GPT-5.4, and the Codex coding agent. According to the AWS Machine Learning Blog, enterprises can now deploy these models in production applications and agents on Bedrock’s high-performance inference engine.

GPT-5.5 serves as the flagship frontier model, designed for complex reasoning tasks, multimodal inputs, and high-stakes enterprise workloads. GPT-5.4 sits as a slightly smaller but still capable alternative, offering faster response times at lower cost for tasks that don’t require maximum capability. The AWS News Blog confirms that both models run on Bedrock’s inference infrastructure.

Codex operates differently from the other two. It functions as an autonomous coding agent rather than a standard text-in, text-out model. Codex can read codebases, write code, run tests, and submit pull requests with minimal human oversight. The agent brings its existing 5-million-user base directly onto AWS infrastructure, as reported by AI Weekly.

Why does this trio matter? Because it covers the full spectrum of enterprise AI needs.

Here’s what each offering targets:

GPT-5.5: Complex reasoning, multimodal analysis, and tasks requiring the highest accuracy
GPT-5.4: Faster, cost-effective inference for production applications at scale
Codex agent: Autonomous software development, code review, and pull request generation
Bedrock integration: All three run through a unified API with consistent billing
Enterprise compliance: Models operate within AWS security boundaries
Multi-model routing: Applications can dynamically select between GPT-5.5 and GPT-5.4 based on task complexity
Production readiness: GA status means SLAs, enterprise support, and guaranteed availability
Vendor diversity: Teams already using Anthropic or Meta models on Bedrock can add OpenAI without changing infrastructure

Model	Type	Primary Use Case	Availability
GPT-5.5	Frontier LLM	Complex reasoning, multimodal tasks	Bedrock GA
GPT-5.4	LLM	Fast, cost-effective inference	Bedrock GA
Codex	Coding agent	Autonomous software development	Bedrock GA

How Does Codex Work on Amazon Bedrock?

Codex on Amazon Bedrock functions as an autonomous coding agent that can independently navigate codebases, write functional code, execute tests, and submit pull requests. Unlike traditional code-completion tools that suggest the next line or function, Codex operates at the task level — you describe what needs to be built or fixed, and the agent plans and executes the full workflow.

The AWS Machine Learning Blog states that Codex is designed for production deployment within enterprise environments. Teams can invoke the agent through the standard Bedrock API, meaning it integrates with existing CI/CD pipelines, code repositories, and development workflows already running on AWS.

When a developer submits a task to Codex, the agent performs several sequential steps. First, it analyzes the relevant codebase to understand context, dependencies, and architecture patterns. Next, it generates the required code changes, following the project’s existing style and conventions. Then it runs automated tests to verify correctness. Finally, it packages the changes into a pull request for human review.

This is not a simple autocomplete tool. It handles multi-file changes.

The AI Weekly report highlights that Codex arrives on AWS with its established 5-million-user base. Those users previously accessed Codex through OpenAI’s own infrastructure. Now they can route requests through AWS, gaining access to Bedrock’s governance controls, audit logging, and enterprise security features.

Key operational details about Codex on Bedrock:

Task-based invocation: Developers describe outcomes, not specific code changes
Multi-file editing: Codex modifies multiple files in a single task execution
Test execution: The agent runs existing test suites to validate its changes
Pull request generation: Completed work appears as standard PRs in connected repositories
API access: Full functionality available through the Bedrock API
Enterprise controls: All actions logged within AWS CloudTrail and governance frameworks
Sandboxed execution: Code runs in isolated environments to prevent unintended side effects
Human review workflow: Pull requests require approval before merging, maintaining oversight

What Security and Compliance Features Does Bedrock Provide?

Amazon Bedrock provides enterprise-grade security, compliance, and governance controls for OpenAI models that would otherwise require custom infrastructure when accessed directly through OpenAI’s API. The AWS News Blog emphasizes that enterprises can now build with OpenAI through the security, compliance, and governance workflows they already have in place on AWS.

Bedrock operates within AWS’s existing compliance framework, which includes SOC 2, ISO 27001, HIPAA eligibility, and FedRAMP authorizations. For enterprises in regulated industries — healthcare, finance, government — this means OpenAI models can be deployed without building separate compliance evidence for a new vendor. The existing AWS artifacts cover the infrastructure layer.

Data handling represents another critical security feature. When enterprises call OpenAI models through Bedrock, prompts and completions remain within the AWS infrastructure boundary. According to the AWS Machine Learning Blog, Bedrock’s inference engine processes requests without storing customer data for model training purposes.

However, there are nuances to understand. The AI Weekly analysis notes that enterprises routing OpenAI calls through AWS should carefully review contractual terms between AWS and OpenAI. Vendor lock-in risk increases if either party alters pricing or access conditions. The convenience of Bedrock comes with dependency on the AWS-OpenAI partnership remaining stable.

Security and compliance capabilities available:

Data residency: Control where inference requests are processed geographically
Private VPC endpoints: Route model traffic through private AWS network connections
AWS CloudTrail logging: Audit every API call to OpenAI models for compliance review
IAM integration: Use existing AWS identity policies to control model access
Encryption: Data encrypted in transit and at rest using AWS KMS
Guardrails for Amazon Bedrock: Apply content filters and topic restrictions to model outputs
Model evaluation tools: Benchmark OpenAI models against other Bedrock providers
Compliance inheritance: Leverage existing AWS compliance certifications

How Do You Access OpenAI Models Through AWS?

Accessing OpenAI models through Amazon Bedrock requires an AWS account with appropriate permissions, model access enabled in the Bedrock console, and API calls routed through the Bedrock inference endpoint. The AWS News Blog provides the official getting-started workflow for enterprises looking to integrate these models.

The first step involves navigating to the Amazon Bedrock console and requesting access to the OpenAI models. AWS requires explicit model access requests because each model provider has separate terms and conditions. Once approved — typically within minutes for standard accounts — the models appear in the Bedrock model catalog alongside offerings from Anthropic, Meta, Cohere, and other providers.

After enabling access, developers can interact with the models through several methods. The Bedrock console provides a playground for testing prompts and comparing outputs. For production applications, the Bedrock API accepts standard HTTP requests. AWS SDKs for Python (boto3), JavaScript, Java, and other languages include Bedrock client methods.

Getting started is straightforward if you already know AWS.

The AWS Machine Learning Blog confirms that all three offerings — GPT-5.5, GPT-5.4, and Codex — follow the same access pattern. This consistency means teams can switch between models or run A/B tests without changing their integration code.

Steps to access OpenAI models on Bedrock:

Open Bedrock console: Navigate to Amazon Bedrock in the AWS Management Console
Request model access: Enable GPT-5.5, GPT-5.4, and Codex in the model access settings
Review terms: Accept the OpenAI provider terms and conditions
Test in playground: Use the Bedrock playground to experiment with prompts
Integrate via API: Call the Bedrock runtime API from your application code
Configure IAM policies: Set up fine-grained access controls for different teams
Monitor with CloudWatch: Track usage, latency, and error rates
Set up guardrails: Apply content safety filters specific to your use case

import boto3
import json

bedrock = boto3.client('bedrock-runtime', region_name='us-east-1')

response = bedrock.invoke_model(
 modelId='openai.gpt-5-5',
 contentType='application/json',
 accept='application/json',
 body=json.dumps({
 'messages': [
 {'role': 'user', 'content': 'Explain quantum computing in 50 words'}
 ],
 'max_tokens': 200
 })
)

result = json.loads(response['body'].read())
print(result['choices'][0]['message']['content'])

Pricing follows the standard Bedrock model — pay per token for inference, with on-demand and provisioned throughput options available. The Digg coverage notes that this launch gives enterprises a new way to build with OpenAI models through infrastructure they already manage.

What Are the Pricing and Cost Implications?

OpenAI models on Amazon Bedrock follow AWS’s standard pay-per-use pricing model, meaning enterprises pay only for the tokens they consume without upfront commitments. According to the AWS News Blog, Bedrock’s high-performance inference engine provides the infrastructure layer, and billing flows through standard AWS accounts rather than requiring a separate OpenAI billing relationship. The exact per-token rates for GPT-5.5 and GPT-5.4 on Bedrock have not been publicly disclosed in the available sources at the time of this writing.

What does this mean for enterprise budgets? Organizations already running workloads on AWS can consolidate their AI spending into a single invoice, reducing administrative overhead associated with managing multiple vendor contracts. For companies with existing AWS Enterprise Discount Programs or committed spend agreements, OpenAI model usage on Bedrock may count toward those commitments, though this depends on specific contractual terms.

The Codex coding agent introduces a different pricing consideration. Since Codex operates as an autonomous coding agent rather than a simple inference endpoint, costs may accrue differently compared to standard chat completion calls. Enterprises should monitor usage closely during initial deployment phases to establish baseline cost patterns. Budget planning requires careful attention here.

Volume discounts through AWS may also apply for high-throughput workloads, similar to how other Bedrock model providers structure their pricing. Organizations should consult their AWS account team for specific rate cards and any available private pricing agreements for OpenAI models on Bedrock.

How Does This Compare to Using OpenAI Directly?

Using OpenAI models through Amazon Bedrock instead of the direct OpenAI API shifts the infrastructure, compliance, and governance responsibilities to AWS’s managed environment. The AWS Machine Learning Blog confirms that enterprises can deploy GPT-5.5, GPT-5.4, and Codex through Bedrock’s high-performance inference engine, which provides a different operational model compared to calling OpenAI’s endpoints directly. The underlying model capabilities remain identical regardless of access path.

Why would an enterprise choose Bedrock over direct API access? The primary differentiator is integration with existing AWS security and compliance workflows. Bedrock provides unified access controls through AWS Identity and Access Management, audit logging through AWS CloudTrail, and encryption through AWS Key Management Service. These are the same governance tools that AWS customers already use for databases, storage, and compute resources.

The AI Weekly report highlights a critical consideration: enterprises routing OpenAI calls through AWS are subject to both AWS and OpenAI contractual terms, which introduces dual-vendor dependency. With direct OpenAI access, organizations manage a single vendor relationship. With Bedrock, they manage two. This affects contract negotiations, SLA enforcement, and dispute resolution processes.

Latency characteristics may also differ between the two access paths. Bedrock’s inference engine routes requests through AWS infrastructure, which could add network hops compared to direct OpenAI API calls. However, AWS’s global edge network and regional deployments may offset this for workloads already running on AWS. Performance testing is essential before committing.

From a feature availability standpoint, the AWS News Blog indicates that GPT-5.5, GPT-5.4, and Codex are all available on Bedrock, suggesting feature parity with the direct API for these specific models. However, new OpenAI features or model updates may not arrive on Bedrock simultaneously with their direct API launch. Lag time varies.

What Vendor Lock-in Risks Should Enterprises Consider?

The AI Weekly coverage explicitly warns that routing OpenAI calls through AWS increases vendor lock-in risk if either party alters pricing or access conditions. This dual-dependency means enterprises are simultaneously locked into AWS infrastructure and OpenAI model capabilities. Codex’s reported 5-million-user base now represents a concentrated dependency on both vendors for coding agent functionality.

Vendor lock-in manifests in several ways when using OpenAI on Bedrock. First, application architectures built around Bedrock’s API structure require engineering effort to migrate to direct OpenAI access or alternative model providers. The API surface differs between Bedrock and OpenAI’s native endpoints, meaning switching costs include code refactoring, testing, and deployment changes.

Second, data residency and processing commitments made through AWS may not transfer seamlessly if an organization decides to move workloads elsewhere. AWS compliance certifications and data handling assurances apply specifically to workloads running on AWS infrastructure. Migrating away means re-establishing compliance posture with a new provider.

Third, pricing changes from either vendor can disrupt cost projections without straightforward alternatives. If OpenAI raises model pricing or AWS adjusts Bedrock hosting fees, enterprises have limited negotiation leverage when locked into both ecosystems simultaneously.

Mitigating lock-in requires architectural discipline. Organizations should abstract model access behind internal APIs, avoid vendor-specific prompt engineering patterns, and maintain evaluation benchmarks across multiple model providers. Building portable evaluation frameworks takes effort but pays dividends.

What Use Cases Benefit Most From This Integration?

The combination of OpenAI’s frontier models with AWS’s enterprise infrastructure creates compelling advantages for specific workload categories. According to the Digg coverage, this integration gives enterprises a way to build on Amazon Bedrock with OpenAI through the security, compliance, and governance workflows they already have in place. Use cases that demand both model capability and enterprise-grade controls benefit most.

Code generation and software development workflows represent the most direct beneficiary, given Codex’s availability as a coding agent on Bedrock. Development teams already using AWS for compute, storage, and deployment can now integrate Codex into their CI/CD pipelines without routing code through external APIs outside their security perimeter. This is particularly valuable for organizations with strict data sovereignty requirements.

Financial services and healthcare organizations operating under regulatory scrutiny gain from Bedrock’s compliance certifications. These industries require audit trails, encryption at rest and in transit, and access controls that AWS provides natively. Running GPT-5.5 for document analysis, risk assessment, or clinical decision support within Bedrock keeps sensitive data within controlled environments.

Customer service automation represents another strong use case. Enterprises running contact center infrastructure on AWS Connect or similar services can integrate GPT-5.5-powered conversational agents with minimal architectural complexity. The unified AWS billing and monitoring stack simplifies operational management.

Internal knowledge management systems also benefit. Companies storing documentation, policies, and institutional knowledge in AWS services like S3 or OpenSearch can build retrieval-augmented generation systems using OpenAI models without moving data outside the AWS ecosystem. Data locality matters for performance and compliance.

How Does This Affect the Competitive Cloud AI Landscape?

The general availability of OpenAI models on AWS Bedrock reshapes the competitive dynamics among cloud providers offering managed AI services. AWS now provides access to what many consider the most capable frontier models alongside its existing model catalog from Anthropic, Meta, Cohere, and other providers. This positions Bedrock as the most comprehensive single platform for accessing diverse foundation models.

Microsoft Azure’s exclusive partnership with OpenAI previously served as a key differentiator for enterprises wanting OpenAI models with enterprise cloud controls. AWS’s addition of OpenAI models to Bedrock neutralizes that advantage, giving AWS customers equivalent access without requiring multi-cloud arrangements. The competitive landscape just shifted significantly.

Google Cloud, which offers its Gemini models through Vertex AI alongside third-party models, now faces pressure to maintain its model provider partnerships. AWS’s ability to attract OpenAI demonstrates that model providers are willing to distribute through multiple cloud platforms, reducing the effectiveness of exclusive cloud-model partnerships as competitive moats.

For Anthropic, which has been a flagship model provider on Bedrock, the addition of OpenAI creates an interesting dynamic. AWS customers can now compare Claude and GPT models side-by-side within the same infrastructure, making direct performance and cost comparisons straightforward. This benefits enterprises but increases competitive pressure on model providers.

Smaller cloud providers and specialized AI platforms face an even steeper competitive challenge. When the two largest cloud providers both offer OpenAI models with enterprise controls, the value proposition of smaller platforms must come from differentiation in specific verticals, pricing, or specialized tooling rather than model access alone.

Frequently Asked Questions

Can I use OpenAI Codex on AWS without an OpenAI account?

Yes, you can use Codex on Amazon Bedrock without maintaining a separate OpenAI account. The AWS News Blog confirms that GPT-5.5, GPT-5.4, and Codex are generally available on Bedrock, with billing and authentication handled entirely through your AWS account. This eliminates the need to manage credentials or payment methods with OpenAI directly.

Which AWS regions support OpenAI models on Bedrock?

The available sources do not specify a complete list of supported AWS regions for OpenAI models on Bedrock at the time of this writing. However, the AWS Machine Learning Blog indicates that models are deployed on Bedrock’s high-performance inference engine, which typically operates across multiple AWS regions. Organizations should check the AWS Bedrock console for current regional availability.

How does Bedrock pricing compare to OpenAI direct API pricing?

Specific per-token pricing comparisons between Bedrock and direct OpenAI API access have not been disclosed in the available sources. The AI Weekly report notes that enterprises are subject to both AWS and OpenAI contractual terms when using Bedrock, which may result in pricing structures that differ from direct API access. AWS account teams can provide specific rate information for OpenAI models on Bedrock.

Is there a free tier for OpenAI models on Amazon Bedrock?

The available sources do not mention a specific free tier for OpenAI models on Amazon Bedrock. AWS typically offers free tiers for some services, but model provider pricing on Bedrock is set independently and may not include free usage allowances. Organizations should consult the AWS Bedrock pricing page or their AWS account representative for current free tier eligibility.

Summary

The general availability of OpenAI GPT-5.5, GPT-5.4, and Codex on Amazon Bedrock represents a significant shift in how enterprises access frontier AI models. Here are the key takeaways:

Dual-vendor dependency requires careful planning. Using OpenAI through AWS means managing relationships with both vendors, and the AI Weekly report warns that pricing or access condition changes from either party could disrupt operations.
Compliance and governance are the primary differentiators. Bedrock provides AWS-native security, audit, and encryption controls that enterprises in regulated industries require, making this integration valuable for financial services, healthcare, and government workloads.
Competitive dynamics have shifted. AWS now matches Microsoft Azure’s access to OpenAI models, reducing the effectiveness of exclusive cloud-model partnerships as competitive differentiators.
Vendor lock-in risk is real but manageable. Organizations should invest in abstraction layers and multi-model evaluation frameworks to maintain flexibility if they need to switch providers.
Codex integration enables new development workflows. With Codex’s reported 5-million-user base now accessible through Bedrock, development teams can integrate AI-powered coding assistance directly into AWS-hosted CI/CD pipelines.

If your organization is evaluating OpenAI models on Bedrock, start with a proof-of-concept deployment targeting a specific workload. Compare performance, cost, and operational overhead against your current approach. The integration is live and generally available — the question is whether it fits your architecture and budget.