RightSignal
2 of 13

Your RightSignal

AI2 min

Microsoft unveils MAI-Base-2, a large AI model it says is far cheaper to train

Microsoft AI announced MAI-Base-2, a base model with 2.8 trillion parameters in total. It is a mixture-of-experts design, meaning only part of it, 80 billion parameters, is active at a time. Microsoft claims it is 4 to 12 times more efficient in pretraining compute than DeepSeek-V4-Pro-Base. It follows MAI-Thinking-1, a reasoning model Microsoft announced in June on top of MAI-Base-1.

Hype check: The efficiency figures are Microsoft's own. It is a base model, so the products built on it matter more to most people, and those come later.
Company claimTechmeme