MAI-1

MAI-1 was reported by The Information in May 2024 as Microsoft's in-house large language model effort, overseen by Mustafa Suleyman after he joined Microsoft as head of consumer AI. The model was said to target roughly 500 billion parameters, far larger than Microsoft's earlier open Phi models, positioning it to compete with Google's Gemini and OpenAI's GPT-4.

MAI-1 was never officially launched: the exact purpose was reportedly undecided, a preview at Microsoft Build was floated but never materialized, and subsequent reporting described underwhelming training runs and internal friction. It remains a historical footnote in the AI arms race, illustrating Microsoft's attempt to build frontier models independent of its OpenAI partnership. This entry is flagged honestly as a reported-but-unreleased project.