Bring more intelligence to everyday work with GPT-6 Sol and GPT-6 Luna on Amazon Bedrock
Amazon Web Services Tanvi Girinath ● Covered by 7 sources
GPT-6 Sol and GPT-6 Luna are now on Amazon Bedrock. AWS is splitting heavy reasoning from high-volume work, and cutting the price from GPT-5.6.
Based on reporting by Amazon Web Services, Tanvi Girinath — read the original for the full story.
Summary, retelling and take written by AI under human oversight; images are AI-generated illustrations. How we work · Report an error
AWS has put GPT-6 Sol and GPT-6 Luna into general availability on Amazon Bedrock, giving developers two new ways to fit model strength to the job instead of forcing every request through the same expensive setup.
That split is the point. GPT-6 Sol is aimed at recurring work that still needs real depth: coding, debugging, refactoring, code review, data analysis, and multistep processes across tools. AWS says OpenAI found it made about half as many factual mistakes as GPT-5.6 Sol on an internal factuality evaluation, and it is meant to carry more of the work from investigation through implementation and validation while explaining what it changed and what it could not confirm.
GPT-6 Luna is the cheaper, tighter option for high-volume chores. It handles extraction, summarization, classification, and focused questions across many users or applications, where every extra token and every bit of latency adds up fast. OpenAI’s evaluations also showed better factual reliability and clearer communication of results for Luna, and AWS says reasoning effort can be adjusted per request to balance quality, responsiveness, and cost.
The other useful detail is that these models do not live in isolation. AWS says a single application can move from Luna to Sol to GPT-6 Astra as the task gets harder, without rebuilding context each time. Both new models support explicit prompt caching on Bedrock, so repeated instructions, tool definitions, policies, and reference material can be reused instead of processed again.
AWS is also leaning hard on the enterprise pitch. Bedrock puts the models behind IAM policies, CloudTrail auditing, VPC endpoints through PrivateLink, and hardware-isolated infrastructure with zero-operator access. AWS says inference data is not used for model training, users do not need to opt into sharing data with OpenAI, and classifier-flagged traffic can be retained by AWS for up to 30 days, with zero data retention available through an AWS account team.
My take — AI-written commentary, not fact-checked reporting
This is the right way to package frontier models: stop pretending one giant model should do every job. The real story is not the shiny name, it’s that AWS is turning AI into something closer to infrastructure, with caching, controls, and model choice instead of demo-room theatrics. That is boring in the best possible way.
Read more about this at: Amazon Web Services