Construct with OpenAI’s smartest mannequin household so far on Amazon Bedrock’s next-generation inference engine.
Organizations scaling autonomous brokers and AI-powered merchandise want frontier intelligence that reliably executes throughout a whole lot of steps, from coding brokers transport product code, to cybersecurity analysis investigating new assault surfaces, to genomics workflows analyzing complete genetic sequences end-to-end. These workloads run on delicate information, demand constant throughput underneath unpredictable masses, and function in environments the place information residency and safety are non-negotiable.
OpenAI’s GPT-5.6 Sol, Terra, and Luna at the moment are typically obtainable on Amazon Bedrock, bringing OpenAI’s smartest mannequin household so far to Amazon Bedrock’s next-generation inference engine constructed for prime efficiency, safety, and reliability. GPT-5.6 units a brand new commonplace for intelligence and effectivity, so you possibly can clear up more durable issues in much less time and with extra intelligence per token. Pricing is per OpenAI’s first-party pricing, and utilization counts towards your present AWS commitments.
Frontier intelligence from core to quick inference
GPT-5.6 introduces a brand new naming system for OpenAI. The numbers determine generations, and Sol, Terra, and Luna determine persistent characteristic layers that may advance at their very own tempo.
- GPT-5.6 Sol is the flagship inference mannequin and probably the most highly effective in OpenAI so far. In accordance with OpenAI, Sol units a brand new state-of-the-art within the Synthetic Analytics Coding Agent Index by 80 factors (2.8 above the following greatest mannequin), whereas utilizing lower than half the output tokens, taking lower than half the time, lowering prices by a couple of third, and scoring 73.5% on cybersecurity analysis’s ExploitBench vs. GPT-5.5 with an equal output token price range. It obtained a rating of 47.9%. Within the Brokers Final Examination, a long-term skilled workflow evaluation throughout 55 disciplines, Sol achieved a brand new excessive of 53.6, beating the following greatest mannequin by 13.1 factors. Even at reasonable inference effort, we nonetheless have an 11.4 level lead at a couple of quarter of the estimated price. Sol additionally introduces most inference effort, which lets you dial up the computing for advanced duties. Use Sol for autonomous coding brokers, vulnerability analysis, drug discovery workflows, and duties that require multi-step deep reasoning.
- GPT-5.6 Terra A well-balanced mannequin excellent for on a regular basis manufacturing work. It gives higher efficiency than GPT-5.5 at a decrease price. Terra can be utilized at no flagship value for code era, content material workflows, structured information extraction, and general-purpose agent duties that require highly effective inference.
- GPT-5.6 Luna It’s a quick and inexpensive mannequin. Use Luna for high-volume inference duties the place per-token latency and price are paramount, reminiscent of classification, summarization, routing, and real-time functions.
Three tiers permit you to appropriately configure mannequin performance and price for every workload. The GPT-5.6 mannequin completes duties with fewer output tokens than earlier variations, leading to higher efficiency per greenback.
An inference engine constructed with scale in thoughts
Agent site visitors is usually bursty. A single consumer request can set off a whole lot of mannequin calls, and demand can change quickly as utilization will increase. Amazon Bedrock’s next-generation inference engine swimming pools capability to soak up demand spikes whereas isolating throughput for every buyer. This reduces the necessity to decide on between shared capability and predictable utility efficiency. Intra-region inference retains requests inside a specified AWS Area and helps groups meet stringent information residency necessities.
Brokers and multi-step workloads additionally repeat a lot of their context between calls. To make the most of that repetition, Amazon Bedrock’s GPT-5.6 introduces immediate caching with specific cache breakpoints. While you mark a cache breakpoint on a reusable portion of a immediate, Amazon Bedrock reuses the processed context for subsequent requests that share it, so every name incurs fees just for new work. Cached inputs are billed at a 90% low cost and may be reused for at the very least half-hour. That is lengthy sufficient to cowl the bursts of calls generated by a single agent execution with out rising prices because the workload grows.
Combining strong security and {hardware} safety
Increased efficiency fashions require stronger safety measures. GPT-5.6 options OpenAI’s most strong security stack so far, fashioned by our most in depth analysis interval so far, combining human purple teaming and in depth automated testing. The security stack contains model-level denials for prohibited actions, real-time misuse classifiers, steady monitoring, and account-level enforcement for persistent patterns.
With Amazon Bedrock, these protections are positioned on high of hardware-level safety. Amazon Bedrock makes use of a chip-enforced Zero Operator Entry (ZOA) safety mannequin, which prevents AWS operators from accessing your prompts and enter. All mannequin calls are executed inside a Digital Personal Cloud (VPC) underneath AWS Identification and Entry Administration (IAM) insurance policies and logged in AWS CloudTrail. Information perimeter insurance policies forestall leakage throughout account and community boundaries. On the mannequin supplier’s request, site visitors information flagged by the classifier is retained for as much as 30 days for automated fraud detection.
Different methods to get GPT-5.6 working with Amazon Bedrock
In parallel with GPT‑5.6, OpenAI launched ChatGPT Work, an agent of ChatGPT for bigger multi-step duties. The up to date ChatGPT desktop app for Mac and Home windows combines chat, work, and Codex into one expertise, with two devoted brokers for various kinds of work.
- work: The brand new Work agent is constructed for bigger, multi-step duties. Accumulate info throughout apps and recordsdata, use the net, create completed supplies like sheets, slides, paperwork, websites, and extra, and hold engaged on advanced initiatives for hours on finish. Customers can monitor progress, change course, and approve vital actions.
- Codex: Codex continues to be a robust coding agent for builders and technical consultants. You may work together with native recordsdata, repositories, terminals, developer instruments, and growth environments to create options, repair bugs, run exams, and open pull requests.
Customers can configure their apps to make use of GPT-5.6 via Amazon Bedrock’s Responses API. Run OpenAI’s smartest household of fashions on Amazon Bedrock and get cutting-edge intelligence with the safety and scale of AWS.
Get began right this moment
GPT-5.6 Sol is accessible within the US East (N. Virginia) and US East (Ohio) AWS Areas. GPT-5.6 Terra and Luna can be found in US East (N. Virginia), US East (Ohio), and US West (Oregon).
Get began utilizing Sol, Terra, and Luna within the Amazon Bedrock Console or programmatically via the Responses API. For extra info, see the Amazon Bedrock documentation and go to the Amazon Bedrock product web page.
In regards to the creator
