Overview

The latest MAI models are trained from scratch with zero distillation on enterprise-grade, commercially licensed data, providing a clean lineage that is deployable into production with confidence. The flagship MAI-Thinking-1, a 35B-parameter reasoning model with a 256K context window, showcases the ability to compete with frontier models on tough benchmarks like SWE-Bench Pro at a fraction of the size and cost. Alongside it sit fast, inference-efficient models for coding and a chart-topping image model that already ranks among the best for editing.

Why Microsoft Models on Foundry

MAI models are coming to Microsoft Foundry, where they run with first-party managed compute, granular quota, and zero data egress. The real differentiator is Microsoft Frontier Tuning: using reinforcement-learning environments, you adapt MAI models to your own data and workflows within your compliance boundary, so the resulting model stays yours.