Microsoft Unveils First In-House Reasoning AI Model, MAI-Thinking-1, to Reduce OpenAI Reliance (MSFT)

Microsoft (MSFT) launched a suite of seven in-house AI models on June 2, 2026, including its first reasoning model, MAI-Thinking-1, at its Build Conference, as the company moves to diversify its AI capabilities and reduce dependence on partner OpenAI. The mid-sized, cost-efficient MAI-Thinking-1 matched Anthropic’s Claude Sonnet Opus 4.6 in coding benchmarks, according to Microsoft. The rollout follows strained relations with OpenAI. In January 2026, the companies revised their cloud agreement, giving Microsoft right of first refusal for capacity. In April, they ended Microsoft’s exclusive access to OpenAI’s intellectual property, permitting OpenAI to sell through competitor platforms like Amazon (AMZN). A restructured revenue-sharing deal capped Microsoft’s payments to OpenAI, while OpenAI’s payments to Microsoft continue. OpenAI is widely expected to pursue an initial public offering later this year, which could deliver a windfall for Microsoft, its early and prominent backer.