Microsoft (MSFT) and OpenAI (OPAI.PVT) were once close pals. But after years of tension between the two, they’re now best described as frienemies.
On Tuesday, Microsoft drilled another hole in the duo’s sinking relationship with the debut of its MAI-Thinking-1 AI model, a midsize model that the company said is intended for high-efficiency, low-token-cost situations.
In terms of capabilities, Microsoft claimed that MAI-Thinking-1 matches Anthropic’s Claude Opus 4.6 on the SWE-Bench Pro coding benchmark. Independent testers, Microsoft said, also preferred its model to Anthropic’s Sonnet 4.6. Anthropic debuted its latest flagship model, Opus 4.8, last week.
Microsoft’s effort is part of its push to decrease its reliance on OpenAI’s technology to power its various Copilot offerings. The Windows developer isn’t ditching OpenAI, though. Instead, the company is continuing to move forward with a model-agnostic approach that allows users to select which AI models they want to tap into, including Anthropic’s Claude.
Microsoft has poured roughly $13 billion into OpenAI over the years. In return, Microsoft signed OpenAI as an exclusive cloud partner and got access to OpenAI’s intellectual property.
But a lot has changed since Microsoft invested an initial $1 billion in the company in 2019. Over the years, OpenAI pushed Microsoft to provide more computing capacity than the software giant was willing to offer.
Microsoft eventually lost its status as OpenAI’s exclusive cloud provider in January 2025 and then lost its right of first refusal to serve OpenAI’s cloud needs in October.
In April, the companies further rewrote their agreement, with Microsoft losing exclusive access to OpenAI’s intellectual property, though it no longer has to make revenue-sharing payments to the company. OpenAI still has to make those payments, but they’re now capped.
OpenAI is expected to go public later this year, putting it up against archrival Anthropic (ANTH.PVT), which has already confidentially filed its IPO paperwork.
