GPT-6 Astra, Sol, and Luna: Tools for Production Agents in Microsoft Foundry

Key Points

GPT-6 Sol and Luna are now generally available in Microsoft Foundry.
Cost per task matters more than price per token for agent ROI.
Standard deployment covers 28 global regions and US and EU Data Zones.

Microsoft Azure has expanded its GPT-6 model family inside Microsoft Foundry. According to the Azure blog, GPT-6 Sol and GPT-6 Luna are now generally available. This follows the earlier launch of GPT-6 Astra. The goal is to give developers better tools for building production AI agents that finish real tasks instead of chattering.

What is changing

The GPT-6 series now offers three distinct tiers for different jobs. Astra handles demanding reasoning and software engineering work. Sol supports complex workflows and coding, while Luna targets extraction and request routing. Teams can pick a model based on workload needs rather than using one size for everything. Each model fits a specific stage in your agent lifecycle.

Input and output prices differ widely by model and context length. Astra charges $50.00 per million output tokens for short context. Luna costs just $0.50 for the same volume. US Data Zone premium adds 10% to Global Standard rates.

Why it matters

This update matters most to cloud architects and developers. They need to split workloads so Luna handles routine volume while Astra handles judgment calls. Migrating agents requires checking tool calls and context limits before swapping models.

Legacy model users should evaluate GPT-5.6 Sol or newer versions. The impact is meaningful for scale, but costs rise with capability. Teams must plan budgets using cost per task, not raw token counts, to see real returns.