Microsoft took a significant step into proprietary AI technology when the company announced an entire portfolio of internal models at its Build 2026 conference. The flagship of that collection is MAI-Thinking-1 — the company's first attempt at an advanced reasoning model.

From OpenAI-dependent to independent player

Until recently, Microsoft's AI efforts have been built almost entirely on technology from OpenAI, a company it has invested heavily in. Last year, however, Microsoft began introducing its own internal models. MAI-Thinking-1 now marks the next phase of that strategy: a fully fledged reasoning model developed by the company itself.

The timing is no coincidence. Microsoft and OpenAI recently renegotiated their partnership agreement, according to The Verge, with an outcome that gives Microsoft greater freedom to operate independently in the AI market.

MAI-Thinking-1 is Microsoft's first statement that advanced AI reasoning can no longer be left solely to its partners.
Microsoft challenges OpenAI with its first advanced reasoning model - Bilde 1

What do we know about the model?

Microsoft characterizes MAI-Thinking-1 as a "medium-sized model" and claims it performs on par with leading competitors on key software development benchmarks. The specific benchmarks referenced, or concrete numerical results, are not provided in available sources — Microsoft's own claims should therefore be treated with caution until independent testing is available.

An intensely competitive market to enter

The market MAI-Thinking-1 is now entering is among the most competitive in the entire AI industry. According to independent benchmark research, the top of the leaderboards in 2026 is dominated by models from Anthropic, OpenAI, and Google.

On SWE-Bench Verified — a key standard measuring the ability to resolve real GitHub issues — Anthropic's best models score above 87 percent, while Google's Gemini 3.1 Pro sits at around 80 percent. OpenAI's GPT-5.5 leads Artificial Analysis's Coding Index with 59.1 points.

Microsoft's claim to "match leading models" is an ambitious statement in this company. The company has not published concrete figures in available source material, and it remains unclear which benchmarks are being used for comparison and under what conditions.

In 2026, launching a reasoning model is not enough — it must prove itself against some of the most powerful AI systems ever built.

What does this mean going forward?

Microsoft's internal model development ambitions are not a technological side project — they are a strategic signal that the company wants to reduce its dependence on a single vendor while simultaneously competing directly in the rapidly growing market for advanced AI in software development.

Whether MAI-Thinking-1 actually delivers on what Microsoft promises will become clear in the weeks and months ahead — once independent evaluators and developers gain access to test the model on their own data and tasks.