What's New

AWS has expanded the availability of OpenAI's GPT-5.4 and GPT-5.5 models to the US East (N. Virginia) Region on Amazon Bedrock. GPT-5.5 is OpenAI's most capable model, targeting advanced coding, research, agentic workflows, and software operation, while GPT-5.4 focuses on frontier reasoning, computer use, and long-context production workloads. Both models are accessible through the Responses API and join a growing portfolio of OpenAI models already available on Bedrock.

How It Works

• Both GPT-5.4 and GPT-5.5 are served through Amazon Bedrock's managed inference infrastructure, allowing access via the Responses API without requiring direct OpenAI account credentials or separate API key management.
• Both models support a 272K-token context window, enabling processing of very long documents, multi-turn conversations, and complex multi-step workflows within a single inference call.
• Input modalities include both text and images, making the models suitable for multimodal tasks such as document analysis, screenshot interpretation, and visual reasoning.
• The Responses API supports server-side and client-side tool calling, allowing models to invoke external tools, APIs, or code interpreters as part of agentic task execution.
• Response streaming is supported, enabling real-time token delivery to end-user applications for lower perceived latency in interactive use cases.
• GPT-5.5 is designed to understand open-ended goals, navigate ambiguity, and complete long-running tasks with minimal orchestration overhead from the developer.
• GPT-5.4 is optimized for production workflows requiring reliable multi-step reasoning, software environment interaction, and output verification across complex business systems.
• Additional OpenAI models available on Bedrock include GPT OSS Safeguard 120B, GPT OSS Safeguard 20B, GPT OSS 120B, and GPT OSS 20B, providing a full-stack option for both inference and safety guardrails.

Why It's Important

• US East (N. Virginia) is AWS's largest and most interconnected region, meaning lower latency for the majority of North American enterprise workloads and broader VPC/networking integration options.
• Enterprises can now access frontier OpenAI models within their existing AWS security perimeter, leveraging IAM, VPC endpoints, AWS PrivateLink, and CloudTrail audit logging without data leaving the AWS ecosystem.
• The 272K-token context window directly addresses enterprise pain points around long-document processing, legal contract review, large codebase analysis, and extended agentic task chains.
• Availability on Bedrock enables unified billing, consolidated cost management, and access to Bedrock features like model evaluation, guardrails, and prompt management alongside OpenAI models.
• The inclusion of OpenAI OSS Safeguard models on the same platform allows teams to pair powerful frontier models with dedicated safety layers, simplifying responsible AI deployment.
• Reduced orchestration requirements in GPT-5.5 lower the engineering burden for building autonomous agents, accelerating time-to-production for complex AI-driven workflows.

How It's Different

• Unlike accessing OpenAI models directly via the OpenAI API, Bedrock hosting means no separate OpenAI account is required, and all traffic, logging, and access control flows through AWS-native tooling.
• The 272K-token context window is substantially larger than many competing models available on Bedrock, supporting use cases that would otherwise require chunking or retrieval augmentation.
• GPT-5.5's explicit design for minimal orchestration differentiates it from models that require heavy prompt engineering or external agent frameworks to complete multi-step tasks reliably.
• GPT-5.4's computer use capability—operating software environments and verifying outputs—goes beyond standard text/code generation, enabling robotic process automation-style workflows driven by LLM reasoning.
• The availability of companion OSS Safeguard models (20B and 120B) on the same platform is a differentiator versus standalone model providers, offering an integrated moderation layer at varying compute cost points.
• Being available in additional AWS regions (not just a single endpoint) provides geographic redundancy and data residency flexibility that direct OpenAI API access does not natively offer.

When to Prefer It

• Prefer GPT-5.5 when building autonomous agents that must handle ambiguous, open-ended goals across long workflows with minimal human-in-the-loop intervention or custom orchestration code.
• Prefer GPT-5.4 for production business applications requiring reliable multi-step reasoning, tool use, and output verification—such as financial analysis pipelines, compliance checks, or ERP integrations.
• Use either model when your organization requires all AI inference to remain within the AWS security boundary for compliance, data residency, or audit trail reasons.
• Choose these models for long-document workflows—such as legal discovery, medical record summarization, or large codebase review—where the 272K-token context window eliminates the need for complex chunking strategies.
• Prefer these models when you need multimodal input (text + images) for tasks like UI testing automation, document digitization, or visual data extraction within an agentic pipeline.
• Use the GPT OSS Safeguard models in conjunction with GPT-5.4/5.5 when deploying customer-facing applications that require content moderation and guardrail enforcement as a separate, auditable layer.
• Choose Bedrock-hosted OpenAI models when you want to consolidate AI spend, monitoring, and governance across multiple model providers (Anthropic, Meta, Mistral, OpenAI) under a single AWS account and billing structure.

Availability

• Status: Generally Available (GA) as of June 11, 2026.
• Regions: US East (N. Virginia) confirmed; the announcement notes this launch expands availability to "additional AWS Regions," implying prior availability in at least one other region.
• Models available: GPT-5.5, GPT-5.4, GPT OSS Safeguard 120B, GPT OSS Safeguard 20B, GPT OSS 120B, and GPT OSS 20B are all listed in the Bedrock OpenAI model catalog.
• API access: Available through the Amazon Bedrock Responses API with support for streaming, server-side tool calling, client-side tool calling, and projects.
• Context window: 272K tokens for both GPT-5.4 and GPT-5.5.
• Input modalities: Text and image inputs supported; output is text.
• Pricing: Not specified in the announcement; pricing is expected to follow Bedrock's standard on-demand token-based pricing model—consult the Bedrock pricing page for current rates.
• Documentation: Model cards available at https://docs.aws.amazon.com/bedrock/latest/userguide/model-cards-openai.html.

Related Resources

• https://docs.aws.amazon.com/bedrock/latest/userguide/model-cards-openai.html