Microsoft unveils MAI-Base-2, a 2.8T-parameter MoE model

Microsoft AI introduced MAI-Base-2, a mixture-of-experts foundation model with 2.8T total and 80B active parameters. The company claims 4x-12x more pretraining FLOPs-efficiency than DeepSeek-V4-Pro-Base. Its predecessor MAI-Base-1 underpinned MAI-Thinking-1, announced in June.

• MoE model with 2.8T total and 80B active parameters
• Claimed 4x-12x pretraining FLOPs-efficiency over DeepSeek-V4-Pro-Base
• Previous MAI-Thinking-1 launched in June on MAI-Base-1