[UPDATE] Microsoft Grows Its MAI Family to Seven Models

When Microsoft first showed off its homegrown models at Build, the story was a reasoning model and a coding model built without OpenAI's help. The fuller picture has now arrived. Microsoft AI has rolled out a family of seven MAI models, extending the work we covered when the company debuted MAI-Thinking-1 and MAI-Code-1. The new lineup reaches well past text into images, speech, and transcription.

What is in the lineup

The seven are MAI-Thinking-1, MAI-Code-1-Flash, MAI-Image-2.5 and its Flash variant, MAI-Transcribe-1.5, and MAI-Voice-2 with its own Flash version. The naming is a mouthful, but the spread is the point: Microsoft now has a model for most of the jobs it used to outsource.

MAI-Thinking-1 is the company's first large reasoning model, built for long, multi-step problems with 35 billion active parameters and a 256,000-token context window. MAI-Code-1-Flash is the lean coding model, roughly 5 billion active parameters, wired directly into GitHub Copilot, VS Code, and the rest of the Microsoft stack. The company pitches it as comparable to a small frontier model but cheaper to run.

Pictures, voices, and transcripts

The image and audio models are the genuinely new ground. MAI-Image-2.5 handles both text-to-image generation and image editing, and Microsoft says it edges past Google's Nano Banana Pro on the public arena leaderboard. MAI-Voice-2 produces natural speech across 15 languages and can mimic a voice from a short sample, which Microsoft pairs with what it calls strong safeguards against misuse.

The boldest claim is for MAI-Transcribe-1.5, which Microsoft describes as the best transcription model in the world, with state-of-the-art accuracy, support for 43 languages, and roughly five times the speed of competing systems. That is a strong claim, and one worth watching as independent testers get their hands on it. Microsoft also paired the launch with a healthcare partnership, naming the Mayo Clinic among early collaborators.

Why it matters

The throughline is independence. Microsoft has spent years as OpenAI's largest backer and its biggest distribution channel, and it still runs OpenAI models across its products. Building a credible in-house family across image, voice, and code gives Redmond a fallback and a bargaining chip. Whether these models match the frontier or simply hold their own on cost, having them in the building changes the math for a company that wants to stop renting every capability it sells.