Model Releases

Microsoft AI details MAI-Base-2 pretraining, 2.8T-parameter MoE model achieves 4.2x to 12.3x the FLOPs efficiency of DeepSeek-V4-Pro-Base

Original title:Frontier Pre-training at Microsoft AI

AI summary

Microsoft AI announced the completion of MAI-Base-2 pretraining, an 80B-active, 2.8T-total-parameter MoE base model that achieves lower BPB than all open-weight models tested on held-out evaluations and is 4.2x to 12.3x more pretraining FLOPs-efficient than DeepSeek-V4-Pro-Base.

Key Takeaways

  • Microsoft AI announced the completion of MAI-Base-2 pretraining,an 80B-active,2.
  • 8T-total-parameter MoE base model that achieves lower BPB than all open-weight models tested on…
  • 2x to 12.
  • 3x more pretraining FLOPs-efficient than DeepSeek-V4-Pro-Base.

Quick Facts

Published
Updated
Source
Microsoft AI:官方博客(RSS)
Category
Model Releases
Topic
Model ReleasesMultimodal
AI score
66
Read the original Read the full article at the original source
Why it matters

官方披露了 MAI-Base-2 的完整预训练方法,读者可以从中了解 scaling ladder、Muon 优化和多轮 dropless MoE 等可借鉴的工程细节。

Topics

Related entities

Source & copyright

This article is based on curated titles and AI summaries from the AIHOT public API, with attribution and links back. Original copyright remains with the original publishers and authors; this site does not mirror full texts.

Original article · AIHOT entry

How to Cite

When citing this article, keep the title, date, and permalink, and credit Orange News as the source.

Orange News: Microsoft AI details MAI-Base-2 pretraining, 2.8T-parameter MoE model achieves 4.2x to 12.3x the FLOPs efficiency of DeepSeek-V4-Pro-Base. October 16, 2026. https://new.nat6.net/en/news/f7ey01hbckuivmlgen9jq0vaf/

View all →