GPT-5.6 family reaches Amazon Bedrock GA
GPT-5.6 Sol, Terra and Luna are now generally available on Amazon Bedrock's Responses API, four days after their public OpenAI launch.
AWS made OpenAI’s GPT-5.6 Sol, Terra, and Luna generally available on Amazon Bedrock on July 13, four days after the models launched publicly through OpenAI directly. Access is through Bedrock’s Responses API on its next-generation inference engine, not the standard Invoke, Converse, or Chat Completions APIs. Pricing matches OpenAI’s first-party rates, and usage counts toward existing AWS spend commitments.
Region support is uneven. Sol is available only in US East (N. Virginia) and US East (Ohio). Terra and Luna add US West (Oregon). AWS is pitching Sol for autonomous coding and agentic workloads, Terra as a cheaper mid-tier for code generation and structured extraction, and Luna for high-volume classification and routing where latency and cost dominate.
Bedrock adds prompt caching with explicit cache breakpoints, a 90 percent discount on cached input tokens, and a “max” reasoning-effort option beyond what OpenAI offers directly. AWS also describes a hardware-enforced, zero-operator-access security model, though it notes traffic flagged by its abuse classifier can be retained up to 30 days, an exception to Bedrock’s standard zero-retention policy.
What it means for operators
This is a distribution change, not a new model: the Sol, Terra, and Luna lineup launched July 9 with the same pricing tiers. What is new is the path to AWS commitment credits and Bedrock’s own caching and security layer, which matters if you already route other workloads through Bedrock, as xAI and NVIDIA models already do.