Microsoft's MAI-Thinking-1 marks shift to in-house reasoning models

Microsoft unveiled MAI-Thinking-1 at Build 2026 on June 2, its first flagship reasoning model developed in-house. The release signals a strategic pivot for the company, which launched its initial proprietary models last year but had previously relied exclusively on OpenAI's infrastructure. Microsoft and OpenAI recently renegotiated their partnership to allow more independent development on both sides.

MAI-Thinking-1 joins a broader slate of new models announced at the conference. Microsoft positioned it as a "flagship" offering, though the company did not disclose parameter counts, training methods, or benchmark comparisons during the Build keynote. The model's name suggests a focus on multi-step reasoning tasks, aligning with the industry's recent emphasis on chain-of-thought and deliberative inference architectures.

Partnership restructuring

The timing follows Microsoft and OpenAI's revised agreement, which loosened exclusivity clauses that had governed their collaboration since 2019. Microsoft retains access to OpenAI's models but now has latitude to ship competing products under its own brand. The company introduced its first in-house models in 2025, though those were smaller-scale offerings aimed at specific enterprise workloads rather than general-purpose reasoning.

MAI-Thinking-1's release suggests Microsoft intends to compete directly in the frontier model space rather than simply reselling OpenAI's technology. The company has not yet published a model card, technical paper, or availability timeline for developers outside its Azure ecosystem.