Amazon Bedrock has added two OpenAI models that divide everyday workloads by complexity. GPT-6 Sol targets recurring but demanding work such as coding, debugging, data analysis and multistep tool use. GPT-6 Luna is aimed at high-volume extraction, summarization, classification and focused questions.
Both are generally available through Bedrock. The models sit below GPT-6 Astra, which remains the family’s option for tasks where maximum reasoning quality matters more than cost. An application can use Luna to route a request, Sol to investigate difficult cases and Astra only when deeper reasoning could change the outcome.
AWS says both models have significantly lower API pricing than their GPT-5.6 counterparts, although the announcement does not provide figures in the article. Reasoning effort can be adjusted per request, and repeated instructions or tool definitions can be reused through context-management features. OpenAI’s internal evaluation found that Sol made about half as many factual mistakes as GPT-5.6 Sol.
Those factuality results come from the model maker and may not transfer to every domain. Teams still need to test accuracy, latency and total retry cost on their own workflows before deciding how to route work among the three GPT-6 tiers.