Microsoft completes MAI-Base-2 pre-training, outperforming open-weight models

AISummary

Microsoft says MAI-Base-2 outperforms every open-weight model it tested on held-out evaluations.

The company says the model is 4-12x more pre-training FLOPs-efficient than DeepSeek-V4-Pro-Base, the strongest open base model, and that it is using those savings to train a larger model.

Mustafa Suleyman says MAI-Base-2 is a 3T-parameter-class base model built entirely in-house, with post-training results to follow.

Post on XView on X

🚨 AI News | TestingCatalog

@testingcatalog

MICROSOFT 🔥: MAI-Base-2 pre-training has been completed; MAI-Base-3 training cycle is already being prepared as well.

"MAI-Base-2 outperforms every open-weight model we tested on our held-out evaluations. It is 4-12x more pre-training FLOPs-efficient than the strongest open base model, DeepSeek-V4-Pro-Base (see Figure 3 below), and we have put that saving into training a larger model."

* Figure 1 shows the performance of MAI-Base-2 and other models with at least one trillion total parameters across a range of pre-training evaluations (lower is better).

Mustafa Suleyman@mustafasuleyman

We now have state-of-the-art pre-trained base models in the 3T-parameter weight class, built entirely in-house at MAI. Post-training results coming soon -- stay tuned! Lots of details in our technical blog post. https://microsoft.ai/news/frontier-pre-training-at-microsoft-ai/

View quoted post on X

Source: 🚨 AI News | TestingCatalog · x.comPublished Oct 15, 2:17 PM ET · added here 2:25 PM ET

Microsoft completes MAI-Base-2 pre-training, outperforming open-weight models

Topics