Get started with OpenAI GPT-5.6 Sol, Terra, and Luna on Amazon Bedrock
AWS Zohreh Norouzi
OpenAI's GPT-5.6 lineup (Sol, Terra, Luna) just landed on Amazon Bedrock as generally available. That means AWS customers can now run OpenAI's newest models through familiar AWS tools, security, and billing.
Based on reporting by AWS, Zohreh Norouzi — 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 and OpenAI just made an unusual pairing official: three flavors of GPT-5.6, code-named Sol, Terra, and Luna, are now generally available on Amazon Bedrock. This isn't a beta or a limited preview. It's a full production release, accessed through a new endpoint called bedrock-mantle, which speaks the OpenAI Responses API natively. For developers who've built agentic coding tools or long-running reasoning pipelines on OpenAI's own API, the pitch is simple: keep your code, swap the base URL, and suddenly you're running inside AWS's security perimeter instead of OpenAI's.
The three models aren't just marketing variants of the same weights. Sol is the heavyweight, tuned for autonomous coding agents, security research, and multi-step scientific reasoning. Terra sits in the middle, built for everyday production traffic where you need a mix of smarts and affordability. Luna is the speed option, meant for high-volume jobs like classification, summarization, and routing where latency and cost matter more than depth. All three share a 272K-token context window, accept text and image input, and support six reasoning-effort settings from none up to max, so switching between them doesn't require rewriting your integration.
What makes this launch interesting isn't just model choice, it's the infrastructure wrapper around it. Every call runs inside your own VPC, gets logged in CloudTrail, and stays within whatever AWS Region you pick, which matters a lot for anyone with data-residency obligations. Pricing mirrors OpenAI's own first-party rates, and usage counts toward existing AWS spend commitments, so there's no new procurement headache. OpenAI also confirms prompts and completions aren't used for training and aren't shared back to them unless a customer opts in, though classifier-flagged traffic can be retained for up to 30 days for abuse detection.
The practical toolkit AWS ships alongside the launch is worth a mention too. Prompt caching, both implicit and explicit, can cut costs sharply for agentic workloads that reuse system instructions or tool definitions across calls, with cached tokens billed at a 90 percent discount. There's also tool-calling support, a revamped project-based console for side-by-side model evaluation, and a direct hookup for OpenAI's Codex coding agent. None of this is revolutionary on its own, but stacking OpenAI's frontier models onto AWS's compliance and cost machinery is a pragmatic move for enterprises that wanted GPT-5.6 without leaving their existing cloud contracts behind.
My take — AI-written commentary, not fact-checked reporting
This is less about GPT-5.6 being some leap forward and more about AWS quietly becoming the toll booth every model provider has to pass through eventually. Enterprises don't pick models in a vacuum, they pick whatever fits inside the compliance stack they already pay for, and OpenAI clearly knows it. Convenient for customers, sure, but it also means the real power in this deal sits with whoever controls the VPC, not whoever trained the weights.
Read more about this at: AWS