OpenAI’s GPT-5.6 Sol, Terra, and Luna are now generally available on Amazon Bedrock, providing organizations with advanced models for complex AI tasks. These models are designed to handle high-performance workloads, including autonomous agents and AI-powered products, with reliable performance across hundreds of steps. They are optimized for environments where data security and residency are critical, and pricing aligns with OpenAI’s first-party rates, with usage counted toward existing AWS commitments.
GPT-5.6 introduces a new naming system, with Sol, Terra, and Luna representing different capability tiers. Sol, the flagship reasoning model, sets a new state of the art on the Artificial Analysis Coding Agent Index at 80 points, outperforming the next-best model by 2.8 points. It also scores 73.5% on ExploitBench for cybersecurity research, compared to 47.9% for GPT-5.5 at a comparable output-token budget. On Agents’ Last Exam, Sol achieves a score of 53.6, surpassing the next-best model by 13.1 points. Terra is the balanced model for everyday production work, delivering superior performance to GPT-5.5 at a lower cost. Luna is the fast and affordable model, ideal for high-volume inference tasks like classification and summarization.
Amazon Bedrock’s next-generation inference engine supports these models, offering scalable performance and security. It pools capacity to handle demand spikes while isolating customer throughput, ensuring predictable application performance. In-Region inference keeps data within specified AWS Regions, meeting strict data-residency requirements. Prompt caching with explicit cache breakpoints helps reduce costs by reusing processed context across similar requests, with cached input billed at a 90 percent discount and available for 30 minutes.
Source: awsml
