OpenAI’s GPT-6 Sol and GPT-6 Luna are now generally available on Amazon Bedrock. The additions give organizations more options for balancing model capability, operating cost and response time across production workloads. Both have significantly lower API pricing than their GPT-5.6 predecessors.
The models complement GPT-6 Astra, which targets ambitious projects where result quality takes priority over cost. GPT-6 Sol is aimed at complex reasoning and coding tasks that teams perform regularly, while GPT-6 Luna targets focused, repetitive work at high volume. Across these workloads, token consumption, retries and latency all affect how economically AI can be deployed.
GPT-6 Sol for recurring complex work
GPT-6 Sol supports feature implementation, debugging, code refactoring and review, data analysis, and multistep workflows involving tools and applications. Improvements over GPT-5.6 Sol in coding and computer use are intended to help it move from investigating a problem to implementing and validating a solution while retaining the context behind its decisions.
OpenAI’s internal factuality evaluation found that GPT-6 Sol produced approximately half as many factual errors as GPT-5.6 Sol. The model also communicates its actions and results more clearly, helping developers distinguish what it changed and verified from what remains unconfirmed or requires human judgment.
For development teams, the relevant economic measure is the overall cost of obtaining a usable result—not just the price of a request. That includes output quality, token use, retries and latency.
GPT-6 Luna for high-volume tasks
GPT-6 Luna is designed for workflows repeated thousands of times a day, where small per-request costs accumulate. Its intended uses include extracting information from large document collections, summarizing incoming content, classifying inputs and answering narrowly scoped questions across users or applications.
OpenAI’s evaluations indicate improved factual reliability and clearer reporting of results. Teams can also set reasoning effort for individual requests to balance quality, responsiveness and cost. These controls matter in document pipelines where extraction, routing and follow-up each add processing steps.
Model selection and reusable context
Applications can assign different models to different stages: GPT-6 Luna for initial classification, GPT-6 Sol for investigating more complicated cases, and GPT-6 Astra when deeper reasoning could meaningfully affect a decision. This approach places more capable processing where it is most useful while managing overall cost and latency.
GPT-6 Sol and GPT-6 Luna support explicit prompt caching on Amazon Bedrock. Developers can designate prompt content for reuse in subsequent calls to the same model, reducing repeated processing of instructions, tool definitions, policies and reference material.
Examples include coding assistants that repeatedly consult repository instructions, support systems using shared policies, and document workflows applying a standard extraction schema.
Production infrastructure and data controls
Amazon Bedrock runs both models on an inference engine designed for high performance, security and reliability at scale. AWS Identity and Access Management (IAM) policies govern model access, while AWS CloudTrail supports auditing every invocation. Virtual private cloud (VPC) endpoints powered by AWS PrivateLink help keep traffic within network boundaries.
Inference uses hardware-isolated infrastructure with zero-operator access, preventing even AWS operators from accessing prompts or completions during inference. Inference data is not used to train models, and customers do not need to opt into sharing data with OpenAI to use GPT-6 Sol or GPT-6 Luna.
For automated abuse detection, AWS retains traffic flagged by classifiers for up to 30 days and processes it programmatically. Customers can request zero data retention through their AWS account team.
Getting started
Both models are accessible through the Amazon Bedrock console and supported Amazon Bedrock APIs. The service documentation provides details on supported AWS Regions, endpoints, APIs, features, inference profiles and pricing. Teams evaluating Bedrock can also contact AWS to discuss their needs.