OpenAI GPT-5.6 Sol, Terra, and Luna are now available on Amazon Bedrock

0 2 5 minutes read

OpenAI GPT-5.6 Sol, Terra, and Luna are now available on Amazon Bedrock

Build with the most intelligent family of models from OpenAI yet, Amazon Bedrock's next-generation indexing engine.

Organizations scaling autonomous agents and AI-enabled products need edge intelligence that works reliably across hundreds of steps, from coding agents that deploy production code to cyber security research that explores novel attack surfaces to genomics workflows that ultimately analyze entire genome sequences. These workloads operate on sensitive data, demand consistent performance under unpredictable load, and operate in environments where data integrity and security are non-negotiable.

Today, OpenAI's GPT-5.6 Sol, Terra, and Luna are commonly available on Amazon Bedrock, bringing the most intelligent model family from OpenAI to Amazon Bedrock's next-generation engine built for high performance, security, and reliability. GPT-5.6 sets a new standard for intelligence and efficiency, so you can solve complex problems in less time and with more intelligence per token. Pricing matches OpenAI's first-party pricing, and usage counts toward your existing AWS commitments.

Frontier intelligence from outstanding thinking to rapid understanding

GPT-5.6 introduces a new generation system from OpenAI: a number identifies a generation, while Sol, Terra, and Luna identify strong power classes that can progress on their own.

GPT-5.6 Sol is the most prominent and powerful logic model from OpenAI to date. According to OpenAI, Sol sets a new state of the art in the Artificial Analysis Coding Agent Index at 80 points (2.8 more than the next best model) while using less than half the output tokens, takes less than half the time, and costs about one third, and scores 73.5% in ExploitBench for cybersecurity research 4% compuss in GPT in Cybersecurity 4% in 59%. output token budget. In the Final Agents Test, the longest-running professional workflow test in 55 fields, the Sol sets a new high of 53.6, surpassing the next model by 13.1 points. In the average thought effort, it still leads by 11.4 points at about a quarter of the estimated cost. Sol also introduces a large effort of thinking, which allows you to drive the computer for complex work. Use Sol for independent coding agents, vulnerability research, drug discovery workflows, and tasks that require multi-step critical thinking.
GPT-5.6 Terra it is a balanced model of everyday production work. It brings superior performance to GPT-5.5 at a lower cost. Use Terra for coding, content workflows, structured data extraction, and general-purpose agent tasks that require robust thinking without a high price tag.
GPT-5.6 Luna it is a fast and affordable model. Use Luna for high-volume tasks such as partitioning, summarization, routing, and real-time applications where latency and cost per token are critical.

With three tiers, you can model the right size and cost for each workload. GPT-5.6 models complete tasks with fewer output tokens than their predecessors, delivering strong performance per dollar.

Indexing engine built for scale

Agent traffic is often explosive: a single user request can trigger hundreds of model calls, and demand can change rapidly as usage grows. Amazon's next-generation Bedrock engine features pooled capacity to absorb spikes while segmenting output for each customer. This reduces the need to choose between shared capacity and predictable application performance. In-region considerations keep requests within the AWS Region you specify, helping teams meet strict data residency requirements.

Agent workloads and multisteps also repeat a large portion of their total between calls. System instructions, tool definitions, and reference files often stay the same while only the latest input changes. GPT-5.6 on Amazon Bedrock introduces faster caching with explicit cache breakpoints to achieve that replication. You mark a reusable piece of information with a cache breakpoint, and Amazon Bedrock reuses the processed context in subsequent requests that share it, so each call only pays for a new operation. Cached entries are charged at a 90% discount and remain available for re-use for at least 30 minutes. This is far enough to cover the burst of calls made by a single agent without incurring costs as the workload scales.

Hard security meets hardware-enforced security

The most skilled models require strong protection. GPT-5.6 features OpenAI's strongest security stack to date, documented by the most extensive testing period yet, including a human red team and extensive automated testing. The security stack includes model-level denial of denied activity, real-time abuse separators, continuous monitoring, and account-level enforcement of ongoing patterns.

In Amazon Bedrock, these protections rest on top of hardware-level security. Amazon Bedrock uses a zero-operator access (ZOA) security model implemented on the chip, so no AWS operators can access your information or complete it. Every model call runs under your AWS Identity and Access Management (IAM) policies, within your virtual private cloud (VPC), and logs into AWS CloudTrail. Data perimeter policies prevent logging across account and network boundaries. As required by the model provider, flagged traffic data will be retained for up to 30 days for automated abuse detection.

Many ways to put GPT-5.6 on Amazon Bedrock to work

Alongside GPT-5.6, OpenAI introduced ChatGPT Work, an agent in ChatGPT for large, multi-step tasks. The updated ChatGPT desktop app for Mac and Windows brings Chat, Work, and Codex together in one place, with two dedicated agents for different types of work:

Job: The new Task Agent is designed for large, multi-step tasks. It can gather information from all applications and files, use the web, create finished objects such as sheets, slides, documents, and sites, and sit with complex projects for hours. Users can track its progress, change direction, and approve important actions.
Codex: Codex remains a powerful code agent for developers and technical professionals. It can work with local files, repositories, terminals, developer tools, and development environments to write features, fix bugs, run tests, and open pull requests.

Users can configure the application to use GPT-5.6 through the Answers API on Amazon Bedrock. Run the most intelligent family of models from OpenAI on Amazon Bedrock and gain edge intelligence with the security and scale of AWS.

Get started today

GPT-5.6 Sol is available in the following AWS regions: US East (N. Virginia) and US East (Ohio). GPT-5.6 Terra and Luna are located in US East (N. Virginia), US East (Ohio), and US West (Oregon).

Run with Sol, Terra, and Luna in the Amazon Bedrock Console or programmatically through the Answers API. To learn more, see the Amazon Bedrock documentation and visit the Amazon Bedrock product page.