Particle.news
Download on the App Store
17 ARTICLES
·
1h ago

Microsoft Debuts First In‑House AI Models, Pushing Copilot Toward Homegrown Tech

The launches signal a move to cost‑efficient, tightly integrated AI that reduces reliance on outside providers.

Satya Nadella, chief executive officer of Microsoft Corp., speaks during the company event on AI technologies in Jakarta, Indonesia, on Tuesday, April 30, 2024.
Image
Microsoft has released two AI models, emancipating itself from an overreliance on OpenAI's technology.
Satya Nadella and Sam Altman on video conference together.

Overview

  • MAI‑Voice‑1, a speech model, is live in Copilot Daily and Podcasts and can generate a minute of audio in under a second on a single GPU, with hands‑on testing available in Copilot Labs.
  • MAI‑1‑preview, Microsoft’s first end‑to‑end foundation text model, is open for public evaluation on LMArena and will roll into select Copilot text scenarios in the coming weeks.
  • Microsoft says MAI‑1‑preview was trained on roughly 15,000 Nvidia H100 GPUs using an efficiency‑focused approach, and it has operational GB200 clusters for the next wave of models.
  • On LMArena, MAI‑1‑preview currently sits around 13th for text tasks, and Microsoft has opened a form for developers to request early access.
  • The shift to in‑house models arrives as multiple outlets report ongoing Microsoft–OpenAI negotiations over an AGI clause and Azure exclusivity, with a SoftBank investment said to hinge on a year‑end agreement.