Microsoft Foundry makes GPT-6 Sol and Luna generally available

Microsoft made GPT-6 Sol and GPT-6 Luna generally available in Microsoft Foundry on September 22, expanding a GPT-6 lineup that already included Astra.

The release gives Foundry customers two lower-cost GPT-6 options for production workloads. Microsoft says Standard deployments for Astra, Sol and Luna are available across 28 Global regions and the US and EU Data Zones. Its model documentation says Sol and Luna do not require a limited-access request, subject to regional availability.

For short-context use under Global Standard pricing, Microsoft lists Sol at $2 per million input tokens, $0.20 for cached input, $2.50 for cached writes and $10 for output. Luna is listed at $0.10, $0.01, $0.125 and $0.50, respectively. The company’s table puts US Data Zone Standard rates 10% above Global Standard and EU Data Zone Standard rates 20% above it for the listed short- and long-context prices.

Microsoft positions Sol as a general-purpose model for production agents, coding, multistep reasoning, long-context analysis and tool-using workflows. It describes Luna as a smaller, faster option for high-volume extraction, summarization, request routing and routine customer interactions. Those capability and cost-per-task descriptions are vendor claims; Microsoft did not publish independently reproducible Foundry workload data with the announcement.

OpenAI’s API changelog recorded both models as released on September 22. OpenAI says Sol and Luna accept text and image inputs and generate text through the Responses and Chat Completions APIs.

Additional capacity options depend on the model and location. Microsoft says Provisioned Throughput is available for Astra and Sol in Global regions and the US and EU Data Zones. Priority Processing for Sol is available in Global regions and the US Data Zone. Microsoft’s regional availability table also lists the Standard footprint.

The announcement includes testimonials from Manus and Wolters Kluwer about Azure model access, support, governance, data residency and production-workflow benefits. Microsoft did not provide independent deployment metrics or case-study methodology for those customer statements.

Microsoft also says Foundry deployments combine model training with prompt and output filters, tool-call protections, prompt-injection mitigation, identity and access controls, Purview policies, evaluation, tracing and monitoring. The announcement does not include independent measurements of the controls’ effectiveness in this deployment.

Sol and Luna follow Microsoft’s earlier GPT-6 Astra general-availability launch in Foundry, making the September 22 release an expansion of the existing lineup rather than Astra’s debut.