OpenAI GPT-5.4 generally available on Amazon Bedrock in AWS GovCloud (US-West)
Government and regulated-industry teams can now run GPT-5.4 agentic workloads inside the GovCloud compliance boundary with full AWS security controls.
View original announcement →Visual Summary
graph TD
A{{GPT-5.4 on Bedrock GovCloud}}:::announced
B((Gov/Regulated Users)):::external
C(Amazon Bedrock):::compute
D([Responses API]):::feature
E([Deep Reasoning & Computer Use]):::feature
F([Data-in-Partition Guarantee]):::feature
G(IAM & CloudTrail):::compute
H(KMS Encryption):::storage
I([Isolated Queues & Durable State]):::feature
B ==>|"invokes"| A
A ==>|"runs on"| C
C -->|"routes via"| D
A -->|"enables"| E
A -->|"enforces"| F
C -->|"governed by"| G
C -->|"encrypted with"| H
C -->|"provides"| I
classDef announced fill:#ff9900,stroke:#ec7211,color:#fff,font-weight:bold
classDef compute fill:#e3f2fd,stroke:#1565c0,color:#1565c0
classDef storage fill:#e8f5e9,stroke:#2e7d32,color:#2e7d32
classDef feature fill:#fff3e0,stroke:#e65100,color:#e65100
classDef external fill:#f5f5f5,stroke:#616161,color:#616161
What's New
OpenAI GPT-5.4 is now generally available on Amazon Bedrock in AWS GovCloud (US-West), extending access to OpenAI's frontier reasoning model to government agencies and regulated-industry customers. The model runs on Bedrock's next-generation inference engine with the full compliance and data-sovereignty guarantees of the GovCloud partition. This follows the broader GA launch of GPT-5.5, GPT-5.4, and Codex on commercial Amazon Bedrock regions announced in June 2026.
How It Works
- GPT-5.4 is accessed via the OpenAI Responses API routed through Bedrock's
bedrock-mantleendpoints, using the OpenAI Python SDK or curl with a Bedrock-issued API key and a GovCloud-specific base URL. - Inference runs on Bedrock's next-generation engine, which assigns each caller an isolated queue with automated capacity management, ensuring predictable throughput even under heavy concurrent load.
- Full request state is captured durably and continuously during execution, so if a node fails or restarts mid-call, the request resumes from its last checkpoint rather than restarting from scratch.
- All data remains within the AWS GovCloud (US-West) partition; prompts and responses are never used to train models and are never shared with OpenAI or any third party.
- Enabling the model in GovCloud requires a two-step process: first accept the EULA in a linked standard AWS account (us-east-1 or us-west-2) via the Console or AWS CLI, then enable the model through the Model Access page in the GovCloud account.
- Every call inherits standard AWS governance controls: IAM permissions, VPC and PrivateLink network isolation, KMS encryption at rest and in transit, and AWS CloudTrail audit logging.
- GPT-5.4 supports native computer-use capabilities alongside deep reasoning for coding, document analysis, and multi-step agentic task execution.
Why It's Important
- Government agencies and regulated industries (defense, intelligence, healthcare, finance) can now leverage a frontier OpenAI model without sacrificing the strict compliance posture required by FedRAMP High, ITAR, and other government frameworks enforced in GovCloud.
- The data-in-partition guarantee means sensitive government data, PII, and classified-adjacent workloads never leave the GovCloud boundary, directly addressing the primary blocker for AI adoption in federal environments.
- Native computer-use and multi-step agentic capabilities unlock automation of complex, multi-system government workflows—such as document processing pipelines, procurement systems, and IT operations—that previously required human-in-the-loop steps.
- Pricing matches OpenAI first-party per-token rates with no additional Bedrock surcharge, making it cost-competitive with direct API access while adding enterprise reliability and compliance overhead at no extra cost.
- Usage counts toward existing AWS Enterprise Discount Program (EDP) commitments, allowing agencies with large AWS spend to offset AI inference costs against existing contracts.
How It's Different
- Unlike calling OpenAI's API directly, GovCloud deployment ensures all inference stays within a US-government-isolated AWS partition, satisfying data residency and sovereignty requirements that commercial regions cannot meet.
- Compared to GPT-5.5 (the higher-tier model), GPT-5.4 is positioned as the best price-performance option for professional workloads, making it suitable for high-volume agentic tasks where cost efficiency matters alongside capability.
- Bedrock's isolated per-caller queue model differs from shared public API infrastructure, providing more predictable latency and throughput guarantees critical for production government systems with SLA requirements.
- The durable mid-execution state capture is unique to Bedrock's inference engine and is not available when calling OpenAI's API directly, making long-running agentic tasks significantly more fault-tolerant.
- Integration with AWS-native security primitives (IAM, KMS, CloudTrail, PrivateLink) provides a unified audit and access-control plane that is absent when using OpenAI's native API, simplifying compliance reporting for government auditors.
- Unlike self-hosted open-source models in GovCloud, GPT-5.4 requires no infrastructure management, GPU provisioning, or model maintenance, reducing operational burden for government IT teams.
When to Prefer It
- Choose GPT-5.4 on Bedrock GovCloud when your workload involves sensitive government data, CUI (Controlled Unclassified Information), or data subject to ITAR, FedRAMP High, or DoD IL requirements that prohibit processing outside the GovCloud partition.
- Use it for high-volume, cost-sensitive agentic pipelines where GPT-5.5's higher capability tier is unnecessary—GPT-5.4 offers frontier-class reasoning at a lower per-token cost, making it the right choice for throughput-intensive automation.
- Prefer it for multi-step document workflows such as contract analysis, regulatory review, or grant processing, where GPT-5.4's long-context reasoning and tool-use capabilities can automate end-to-end pipelines within a compliant boundary.
- It is the right choice for software development and DevSecOps automation in classified or regulated environments, where GPT-5.4's coding and computer-use capabilities can accelerate delivery without exposing code to commercial infrastructure.
- Select it when you need fault-tolerant agentic tasks that may run for extended periods—Bedrock's durable state capture ensures long-running jobs survive infrastructure failures without manual restart logic.
- Use it when your organization needs a single, auditable AI inference platform that integrates with existing AWS IAM roles, VPC configurations, and CloudTrail logging rather than managing a separate OpenAI API key and audit trail.
Availability
- Status: Generally Available (GA) as of June 3, 2026.
- GovCloud Region: AWS GovCloud (US-West) — the primary supported region for this announcement.
- Commercial Regions: GPT-5.4 is also available in commercial AWS regions; see the AWS Regions page for the full regional compatibility matrix.
- Inference Routing Options: In-Region (data stays in one region), Geographic/Geo (routes within a defined geography), and Global (routes across all commercial regions) — GovCloud workloads use In-Region routing to maintain partition isolation.
- Pricing: Per-token pricing matching OpenAI first-party rates; no additional Bedrock surcharge; usage counts toward existing AWS EDP commitments.
- Access Prerequisite: EULA must be accepted in a linked standard AWS account (us-east-1 or us-west-2) before the model can be enabled in the GovCloud account via the Model Access page.
- Data Handling: Prompts and responses are never used to train models and are never shared with OpenAI; all data remains within the GovCloud partition.
- API Compatibility: Accessible via the OpenAI Responses API using the OpenAI Python SDK or curl, pointed at Bedrock's
bedrock-mantleendpoint.
Related Resources
- https://docs.aws.amazon.com/bedrock/latest/userguide/models-region-compatibility.html
- https://aws.amazon.com/blogs/machine-learning/openai-models-and-codex-on-amazon-bedrock-are-now-generally-available
- https://docs.aws.amazon.com/bedrock/latest/userguide/model-cards-openai.html
- https://aws.amazon.com/blogs/aws/get-started-with-openai-gpt-5-5-gpt-5-4-models-and-codex-on-amazon-bedrock
