Construct with the neatest household of fashions from OpenAI but, on Amazon Bedrock’s next-generation inference engine.
Organizations scaling autonomous brokers and AI-powered merchandise want frontier intelligence that performs reliably throughout lots of of steps, from coding brokers delivery manufacturing code to cyber safety analysis probing novel assault surfaces to genomics workflows analyzing total gene sequences end-to-end. These workloads run on delicate knowledge, demand constant throughput below unpredictable load, and function in environments the place knowledge residency and safety are non-negotiable.
As we speak, GPT-5.6 Sol, Terra, and Luna from OpenAI are typically obtainable on Amazon Bedrock, bringing the neatest household of fashions from OpenAI but to Amazon Bedrock’s next-generation inference engine constructed for high-performance, safety and reliability. GPT-5.6 units a brand new normal for intelligence and effectivity, so you’ll be able to resolve tougher issues in much less time and with extra intelligence per token. Pricing matches OpenAI first-party charges, and utilization counts towards your present AWS commitments.
Frontier intelligence from flagship reasoning to quick inference
GPT-5.6 introduces a brand new naming system from OpenAI: the quantity identifies the era, whereas Sol, Terra, and Luna determine sturdy functionality tiers that may advance on their very own cadence.
- GPT-5.6 Sol is the flagship reasoning mannequin and essentially the most highly effective from OpenAI to this point. In response to OpenAI, Sol units a brand new state-of-the-art on the Synthetic Evaluation Coding Agent Index at 80 factors (2.8 above the next-best mannequin) whereas utilizing lower than half the output tokens, taking lower than half the time, and costing about one-third much less, and scores 73.5% on ExploitBench for cybersecurity analysis versus 47.9% for GPT-5.5 at a comparable output-token funds. On Brokers’ Final Examination, an analysis of long-running skilled workflows throughout 55 fields, Sol units a brand new excessive of 53.6, outperforming the next-best mannequin by 13.1 factors. At medium reasoning effort, it nonetheless leads by 11.4 factors at roughly one-quarter the estimated price. Sol additionally introduces max reasoning effort, which helps you to dial up compute for complicated work. Use Sol for autonomous coding brokers, vulnerability analysis, drug discovery workflows, and duties that require deep multi-step reasoning.
- GPT-5.6 Terra is the balanced mannequin for on a regular basis manufacturing work. It delivers superior efficiency to GPT-5.5 at a decrease price. Use Terra for code era, content material workflows, structured knowledge extraction, and general-purpose agentic duties that want sturdy reasoning with out flagship pricing.
- GPT-5.6 Luna is the quick and reasonably priced mannequin. Use Luna for high-volume inference duties like classification, summarization, routing, and real-time functions the place latency and value per token matter most.
With the three tiers, you’ll be able to right-size mannequin functionality and value to every workload. GPT-5.6 fashions full duties with fewer output tokens than their predecessors, delivering stronger efficiency per greenback.
An inference engine constructed for scale
Agent visitors is commonly bursty: one consumer request can set off lots of of mannequin calls, and demand can change shortly as utilization grows. Amazon Bedrock’s next-generation inference engine swimming pools capability to soak up demand spikes whereas isolating every buyer’s throughput. This reduces the necessity to decide on between shared capability and predictable software efficiency. In-Area inference retains requests throughout the AWS Area that you just specify, serving to groups meet strict data-residency necessities.
Agentic and multi-step workloads additionally repeat a lot of their context between calls. System directions, instrument definitions, and reference recordsdata usually keep the identical whereas solely the most recent enter adjustments. GPT-5.6 on Amazon Bedrock introduces immediate caching with express cache breakpoints to make the most of that repetition. You mark the reusable a part of a immediate with a cache breakpoint, and Amazon Bedrock reuses the processed context on subsequent requests that share it, so every name pays just for the brand new work. Cached enter is billed at a 90 % low cost and stays obtainable for reuse for at the very least half-hour. That is lengthy sufficient to cowl the burst of calls a single agent run generates with out compounding price as workloads scale.
Sturdy security meets hardware-enforced safety
Extra succesful fashions require stronger safeguards. GPT-5.6 options OpenAI’s most sturdy security stack to this point, formed by their most in depth analysis interval but, combining human purple teaming with large-scale automated testing. The security stack contains model-level refusals for prohibited exercise, real-time misuse classifiers, steady monitoring, and account-level enforcement for persistent patterns.
On Amazon Bedrock, these protections sit on prime of hardware-level safety. Amazon Bedrock makes use of a zero-operator entry (ZOA) safety mannequin enforced on the chip, so no AWS operators can entry your prompts or completions. Each mannequin name runs below your AWS Id and Entry Administration (IAM) insurance policies, inside your digital non-public cloud (VPC), and is logged in AWS CloudTrail. Knowledge perimeter insurance policies forestall exfiltration throughout account and community boundaries. As required by the model-provider, classifier-flagged visitors knowledge will probably be retained for as much as 30 days for automated abuse detection.
Extra methods to place GPT-5.6 on Amazon Bedrock to work
Alongside GPT‑5.6, OpenAI launched ChatGPT Work, an agent in ChatGPT for bigger, multi-step duties. The up to date ChatGPT desktop app for Mac and Home windows brings Chat, Work, and Codex collectively in a single expertise, with two devoted brokers for completely different varieties of labor:
- Work: The brand new Work agent is constructed for greater, multi-step duties. It could actually collect data throughout apps and recordsdata, use the online, create completed supplies akin to sheets, slides, docs, and Websites, and stick with complicated tasks for hours. Customers can comply with its progress, change course, and approve essential actions.
- Codex: Codex stays the highly effective coding agent for builders and technical professionals. It could actually work with native recordsdata, repositories, terminals, developer instruments, and growth environments to put in writing options, repair bugs, run exams, and open pull requests.
Customers can configure the app to make use of GPT-5.6 by means of the Responses API on Amazon Bedrock. Run the neatest household of fashions from OpenAI on Amazon Bedrock and get frontier intelligence with the safety and scale of AWS.
Get began at the moment
GPT-5.6 Sol is out there within the following AWS Areas: US East (N. Virginia) and US East (Ohio). GPT-5.6 Terra and Luna can be found in US East (N. Virginia), US East (Ohio), and US West (Oregon).
Get began with Sol, Terra, and Luna within the Amazon Bedrock Console or programmatically by means of the Responses API. To study extra, see the Amazon Bedrock documentation and go to the Amazon Bedrock product web page.
In regards to the authors
