On November 4, 2026, OpenAI announced that eligible customers can use GPT-5.6 Cyber through Daybreak Red on Amazon Bedrock in AWS GovCloud (US-East). Access takes two approvals: OpenAI must approve the customer for Trusted Access for Cyber, then AWS must enable the account. OpenAI’s announcement describes the GovCloud availability; AWS details the account-access step in its model card.
OpenAI had announced Daybreak models on AWS on August 11, 2026. The November 4 announcement covered the GovCloud region specifically. (OpenAI’s earlier announcement)
Access requires OpenAI approval and AWS provisioning
An organization first needs approval through OpenAI’s Trusted Access for Cyber program. After approval, it must ask its AWS account team to enable access in the AWS account. OpenAI approval alone does not activate the model in Amazon Bedrock.
AWS limits inference to listed regions and modes
AWS lists In-Region inference for us-gov-east-1 (AWS GovCloud (US-East)) and us-east-2 (Ohio). It does not support Geo or Global cross-Region inference for this model. The model ID is openai.gpt-5.6-cyber, and its context window is 1,050,000 tokens.
AWS supports the bedrock-runtime and bedrock-mantle endpoints. With bedrock-mantle, the documented API paths are /openai/v1/responses for Responses and /openai/v1/chat/completions for Chat Completions. AWS lists the Standard service tier; Priority and Flex are unsupported.
AWS publishes model-specific FedRAMP Class D and DoD CSP SRG IL-4/IL-5 status separately from its regional availability information. Its GovCloud guide directs customers to that separate certification-status listing.
The model is intended for authorized security work
OpenAI identifies vulnerability research, exploit validation, and security testing as intended work. AWS lists vulnerability research, exploit reproduction, and mitigation development. OpenAI says qualified security practitioners should review findings and proposed fixes before implementation.
OpenAI describes Daybreak Red as allowing more latitude for authorized cybersecurity tasks, with stronger identity verification, monitoring, and access controls. The company also says inference data on Amazon Bedrock is not used to train models and customers do not need to share Daybreak Red data with OpenAI.
AWS token rates depend on region
AWS lists these Standard-tier In-Region rates for short-context requests of 272,000 input tokens or fewer. Each amount is in USD per 1 million tokens.
| Usage | AWS GovCloud (US-East) | Commercial AWS In-Region |
| Input | $16.875 | $13.75 |
| 30-minute cache write | $21.09375 | $17.1875 |
| Cache read | $1.6875 | $1.375 |
| Output | $101.25 | $82.50 |
The cache-write rate applies to each million tokens written to the prompt cache; 30 minutes is how long a cached prompt prefix is retained. For matching input and output rates, GovCloud is about 22.7% more expensive than commercial In-Region. AWS says the commercial In-Region rates already include its 10% fee, so that fee should not be added again. The 1,050,000-token context window is the model’s limit; the listed prices apply only to requests in the short-context band.