As reported by AWS Machine Learning, GPT-6 Sol and GPT-6 Luna are now available on the Amazon Bedrock platform, giving users more options to match intelligence and efficiency in every task. The value of AI at scale depends on two factors: what the model can do and how often it can be used. Greater intelligence expands the complexity of problems the model can solve, from subtle coding problems to multi-step processes using a variety of tools. Efficiency determines how broadly this intelligence can support everyday activities and repeatable tasks, where each additional token, retry, and second of delay is multiplied by the number of requests. GPT-6 Astra sets the upper limit of the GPT-6 family for the most ambitious projects, where obtaining the highest quality results is more important than cost. Organizations also require advanced intelligence capabilities for routine tasks that support products and operations. GPT-6 Sol offers strong reasoning and coding capabilities for complex tasks performed throughout the week, with e
economics suitable for regular use. GPT-6 Luna makes focused, repeatable workloads practical at high volumes, where small differences in latency and cost are multiplied by the number of requests. OpenAI's GPT-6 Sol and GPT-6 Luna are now available on Amazon Bedrock, running on a high-performance inference engine that delivers security and reliability at scale. Both models are offered at significantly lower API prices than their GPT-5.6 predecessors, giving you more opportunity to bring GPT-6 intelligence into production with the performance, control and flexibility you need for your applications. GPT-6 Sol is designed for demanding tasks that are repeated throughout development and operations. It can implement functions, debug problems, refactor and review code, analyze data, and complete multi-step processes using a variety of tools and applications. Improvements over GPT-5.6 Sol in coding and machine usage help it perform tasks from exploration to implementation to validation while maintaining the context of its decisions. In an internal evaluation of actual accuracy, OpenAI found that GPT-6 So
l makes about half as many actual errors as GPT-5.6 Sol. GPT-6 Sol also benefits from clearer communication about its work and results, which helps teams quickly identify gaps and understand where human judgment is still required. Together, stronger execution and clearer reporting make GPT-6 Sol practical throughout the development cycle. The relevant measure here is the total cost of achieving the output being used, including output quality, token usage, retries, and latency. When a task is executed thousands of times a day, the economics of each call determine whether the workflow will scale. A single classification or summary query is inexpensive on its own, but the cost of retrieving, routing, and following up on a complete document accumulates with each additional query. The GPT-6 Luna is designed for workloads where this volume matters. You can use it to extract information from large collections of documents, summarize incoming materials, categorize input data, and answer focused questions from many users or applications. Efficiency
and high volumes also requires consistent results. OpenAI's evaluations show improvements in the actual reliability of GPT-6 Luna and clearer communication of results. You can also adjust the reasoning effort on a per-request basis to balance the quality, responsiveness, and cost required for each task. A single application may require different levels of intelligence as the request progresses. You can use GPT-6 Luna to classify incoming requests, GPT-6 Sol to investigate complex cases, and GPT-6 Astra when additional depth of reasoning can make a significant difference to the decision. This focuses intelligence where it creates the most value, managing latency and cost across the system. At each stage, repeated calls to the same model can reuse instructions, tool definitions, policies, and reference materials. Reprocessing this context may undermine the efficiency achieved by selecting the appropriate model. GPT-6 Sol and GPT-6 Luna support explicit hint caching on Amazon Bedrock. You can mark the contents of a tooltip for reuse, allowing subsequent requests to
focus on new input. This is useful for coding assistants that reuse repository instructions, support applications based on the same policies, and document processes that apply a consistent retrieval scheme. With the increasing use of AI, the quality of the model is only part of what determines whether an application will be successful in production. Teams also need infrastructure that maintains performance as demand changes, economics that persist across repeated requests, and controls that protect sensitive data. Amazon Bedrock provides this foundation for GPT-6 Sol and GPT-6 Luna through a high-performance inference engine built for security and reliability at scale. You can control access to the model through AWS Identity and Access Management (IAM) policies and validate each invocation through AWS CloudTrail. Virtual private network (VPC) endpoints running on AWS PrivateLink help keep traffic within the boundaries of your network. Output occurs on an isolated infrastructure with zero operator access, so even AWS operators cannot access your
hints or results during output. Your output data is not used to train the model, and use of GPT-6 Sol and GPT-6 Luna does not require you to agree to share your data with OpenAI. For automated abuse detection, traffic flagged by the classifier is stored by AWS for up to 30 days and processed programmatically. You can request zero data retention through your AWS team. You can get started with GPT-6 Sol and GPT-6 Luna in the Amazon Bedrock console or programmatically through supported Amazon Bedrock APIs. For information about supported AWS regions, endpoints, APIs, features, output profiles, and pricing, see Amazon Bedrock documentation. The material was prepared by a neural network based on the original publication. AI may make mistakes in details (numbers, names, dates, quotes) - for important information, check with the original source. AWS Machine Learning Introduces Agent Skill Assessment with Strands Evals OpenAI Introduces GPT-6 Sol and Luna with Low Cost and Less Errors Kyutai has released a Voice of Reason model for verbally solving math problems. Nokia opens access to AnyJev, a library for decision-making without training. OpenAI pr
introduced GPT-6 Sol and Luna with new API prices, reducing the cost by 50%.
