GPT-6 Astra, Sol, and Luna: For production agents in Microsoft Foundry

We’re excited to announce the expansion of our GPT-6 lineup, introducing GPT-6 Sol and GPT-6 Luna to the offerings available via the Microsoft Foundry. This new release builds upon the remarkable success of GPT-5.6 Sol and GPT-6 Astra, as we strive to enhance capabilities in Microsoft Foundry. Our aim is to create solutions that minimise noise while providing efficient and effective task completion with agents.

Astra offers improved reasoning and software engineering, making it ideal for challenging tasks that require careful judgement. Azure users are already noticing significant advancements in performance and cost effectiveness as the model requires fewer but more valuable tokens to guide agents.

Completing our line-up, GPT-6 Sol is perfect for general-purpose applications, whereas Luna excels in managing large-scale data and preparatory tasks.

Choose the Right Intelligence for Your Agents

Selecting the ideal model for any task starts with thorough evaluations. Different agents, like those dealing with complex business decisions versus those managing routine queries, have unique requirements. Microsoft suggests beginning with GPT-6 Astra for more intricate tasks. For higher-volume operations, GPT-6 Sol and Luna provide a balanced alternative tailored for production and scalability.

Using GPT-6 Sol for Production AI Agents and Complex Workflows

GPT-6 Sol, alongside its trusted predecessor GPT-5.6 Sol, presents a more economical intelligence option with top-tier efficiency. They’re designed to support enterprise agents, coding, and complex knowledge work—covering tasks that require multi-step reasoning, lengthy context analysis, and workflows utilising various tools. If your team is considering new production tasks or moving away from an older model, Sol is an excellent choice to begin with.

Embrace GPT-6 Luna for Efficient, High-Volume AI Workloads

GPT-6 Luna is the agile cousin of Sol, specifically crafted for high-volume tasks. It’s ideal for activities like data extraction, summarisation, routing requests, and handling regular customer interactions. Save the in-depth reasoning for when it’s truly necessary; there’s no need to deploy the same model for every single job.

As the GPT-6 range evolves, it’s not just about choosing newer models; it’s about enhancing what your agents can achieve while also saving costs. To truly understand your return on investment in AI, customers are encouraged to focus on cost per task rather than simply the price per token.

The accompanying chart illustrates the reasons why enterprise users on Microsoft Foundry are switching to GPT-5.6 Sol and the latest GPT-6 models.

Foundry integrates evaluation and monitoring to help teams make informed decisions. Ultimately, the true measure of progress lies in what customers achieve in a production setting. That’s why Foundry has always advocated for model selection and a flexible, interoperable framework.

The Foundry Edge: What Our Customers Say

Having access to cutting-edge models is just the beginning. Foundry pairs GPT-6’s intelligence with extensive deployment options, catering to the vast needs of enterprise production. As of now, Standard deployment is available for Astra, Sol, and Luna across all 28 Global regions, as well as US and EU Data Zones; Provisioned Throughput for Astra and Sol across Global regions and US and EU Data Zones; and Priority Processing for Sol across Global regions and US Data Zones. The extensive capability and performance of Azure is why OpenAI continues to debut its models on Azure, and why sophisticated clients such as Manus choose Foundry.

The Azure OpenAI models form the core intelligence for Manus. Through Azure, we reliably integrate advanced models into our workflows, enabling us to understand user intent, plan tasks, and execute complex projects. The responsive technical support from Microsoft and quick access to new model capabilities allow us to iterate swiftly and deliver an exceptional AI experience.

—Tao Zhang, Co-Founder & Product Partner, Manus

For those starting their AI journey with Azure: opt for Global for flexible, pay-per-token capabilities, or choose supported Data Zone deployments to meet processing-location requirements. Priority Processing offers a premium lane for responsive, pay-as-you-go experiences, with Provisioned Throughput ensuring reserved capacity and lower latency for crucial production needs. Match the service option to your workload, whether it’s interactive agents or high-throughput business processes.

This is the Foundry advantage: it’s not solely about cutting-edge intelligence but also about the platform that empowers you to leverage it. Teams can align each workload with the right model, deployment choice, and controls, balancing capability, responsiveness, and costs as their usage expands. By integrating these options on Azure, Foundry enables customers to focus on delivering real business value, providing the operational backbone to transition from concept to large-scale production.

Our clients operate in sectors where just finding an answer isn’t enough; it must be the correct one and withstand scrutiny. The latest Azure OpenAI models can methodically reason through issues, a feature vital for the research and compliance tasks our professionals rely on. Employing Microsoft Foundry allows us to implement those workflows into production, using infrastructure and services that are already in line with our governance, data residency, and security needs.

—Brian Diffin, CTO of Wolters Kluwer Tax & Accounting

GPT-6 Pricing and Deployment Options

**Prices may differ for Provisioned Throughput and Priority Processing depending on deployment type. U.S. Data Zone incurs a 10% surcharge compared to global pricing. For up-to-date rates and terms, please visit the Azure OpenAI pricing page.

Create Safer AI Agents with Microsoft Foundry

GPT-6 models operating on Azure come with multiple safety and security layers designed to protect the models and their interactions. The core model incorporates alignment and safety training, while the surrounding prompts and outputs are safeguarded by content filters and guardrails that determine what agents can say. Additionally, tool interactions and responses are governed by controls and prompt injection protections that regulate what actions an agent can undertake. Identity and access are managed according to enterprise policies that dictate reachable data.

Foundry ensures that teams can consistently fortify safety measures as risks change. It implements guardrails at essential checkpoints, including those for prompts, outputs, tool interactions, and responses. Identity and access controls govern agents’ capabilities and access. Microsoft Purview enforces enterprise data policies, while evaluation, tracking, and monitoring provide the evidence needed to enhance those controls over time, with human checkpoints throughout the process.

Transition to GPT-6: Craft Your Next Generation of Agents.

Start building your future agent workloads with Microsoft Foundry. Use GPT-6 Astra for demanding reasoning tasks, GPT-6 Sol for general production needs, and scale your high-volume assignments with GPT-6 Luna. For users of older models, we suggest considering an upgrade to GPT-5.6 Sol or newer.

Your next agent needs more than just a powerful model. Foundry combines an open intelligence stack, flexible deployment options, and Azure enterprise controls to provide you with the confidence to build and scale from your very first workload to full-scale production.

FAQ

1. What is GPT-6 and how does it differ from previous versions?

GPT-6 is an advanced AI model series designed to enhance performance and efficiency in various tasks. It builds on the capabilities of earlier versions like GPT-5, offering improved reasoning, task completion, and overall cost-effectiveness.

2. How can I choose the right GPT-6 model for my project?

Evaluation is key. Consider the complexity of your tasks; for demanding needs, start with GPT-6 Astra. For general production, GPT-6 Sol is ideal, while GPT-6 Luna is suitable for high-volume tasks.

3. What are the deployment options available with Microsoft Foundry?

Microsoft Foundry offers standard deployment options across global regions, along with specific provisions for US and EU Data Zones. Priority processing options are also available depending on your operational needs.

4. How is safety ensured with GPT-6 models?

Safety is embedded in GPT-6 models through multiple layers of training and built-in filters. Guardrails are integrated to monitor interactions, ensuring compliance with enterprise policies and safeguarding against risks.