Changes confirmed medium confidence

Amazon Adds GPT-6 Sol and Luna to Bedrock

The two OpenAI models are generally available through Amazon Bedrock, with prompt caching and AWS access controls but unresolved region-by-region availability.

Edited by Tyronne Panaino

Amazon Web Services made OpenAI's GPT-6 Sol and GPT-6 Luna generally available on Amazon Bedrock on September 22. The AWS availability notice says customers can use the models through the Bedrock console or supported Bedrock APIs.

For teams that already operate AI workloads on AWS, the material change is the deployment surface rather than a new model launch. Sol and Luna can now be evaluated inside Bedrock's managed access path, but AWS directs customers to separate documentation for supported Regions, endpoints, inference profiles, features and pricing. Availability should therefore be checked against the exact Region and interface a workload needs.

Two models for different workload shapes

AWS positions GPT-6 Sol for recurring complex work, including software development, data analysis and multistep tool use. It positions GPT-6 Luna for higher-volume, focused tasks such as summarization, extraction, classification and routing. The announcement says both models support context windows of up to one million tokens, while Luna also exposes adjustable reasoning effort.

Those descriptions are product guidance, not independent evidence that either model is the best choice for a particular application. A useful evaluation should compare task completion, retries, latency and total token use on the same workload. The narrower supported conclusion is that Bedrock now offers two GPT-6 options aimed at different operating profiles.

Prompt reuse is part of the Bedrock offer

The more detailed AWS launch article says both models support explicit prompt caching. Developers can mark reusable material so later requests process the new input rather than the whole repeated prefix again. AWS gives repository instructions, support policies and extraction schemas as examples of content that may be reused.

Prompt caching can reduce repeated processing, but the source does not establish a universal cost or latency saving. Whether it helps depends on how much stable context a workload repeats, how often calls reuse it and which Bedrock pricing and cache rules apply in the selected Region.

The governance boundary is AWS-specific

AWS says model access can be governed with Identity and Access Management policies, invocations can be audited through CloudTrail and VPC endpoints using PrivateLink can keep traffic inside a customer's network boundary. The company also says inference data is not used for model training and customers do not have to opt into sharing that data with OpenAI.

There is an important retention qualifier. AWS says traffic flagged by its automated abuse-detection classifiers may be retained for up to 30 days and processed programmatically. Customers can request zero data retention through their AWS account team. These are AWS's descriptions of its controls and data path; this article does not treat them as an independent security or privacy audit.

What remains to verify

The announcement does not list every supported Region, endpoint, inference profile or price in the fetched page. Teams should verify those details in the current Bedrock documentation and test IAM, logging, network and retention settings in the account that will run the workload. Independent performance and reliability evidence is also still needed before turning AWS's model positioning into a production expectation.

Status

Confirmed. AWS has announced general availability of GPT-6 Sol and GPT-6 Luna on Amazon Bedrock. Internal confidence is medium because availability and platform-control claims come from AWS and OpenAI participants without independent production testing in the fetched evidence.

Sources

Update note: Last reviewed 2026-09-25. We will revise this post if AWS changes regional availability, retention terms or the supported Bedrock interfaces.

Sources

Drafted with AI assistance from source briefs; reviewed for citation completeness and label accuracy.

More Changes coverage