Microsoft finishes pre-training 2.8T-parameter MAI-Base-2

Name: Microsoft finishes pre-training 2.8T-parameter MAI-Base-2
Uploaded: 2026-10-16T00:00:00+05:30
Description: Microsoft AI's new in-house MoE base model (80B active, 2.8T total) beats every open-weight model it tested on its own held-out evals and is 4-12x more compute-efficient than DeepSeek-V4-Pro-Base, per Microsoft; it is now in post-training.

Why it matters

Microsoft AI's new in-house MoE base model (80B active, 2.8T total) beats every open-weight model it tested on its own held-out evals and is 4-12x more compute-efficient than DeepSeek-V4-Pro-Base, per Microsoft; it is now in post-training.

Read the source: Microsoft AI (opens in a new tab)

Watch on YouTube Watch on Instagram

All hits from 16 Oct 2026 More Models Get the daily hits by email