TLDR Tech / Launches
Microsoft AI completes pre-training for new 80B parameter model
Microsoft AI has finished the initial training phase for MAI-Base-2, a new foundation model employing a mixture-of-experts architecture. This model boasts 80 billion active parameters and 2.8 trillion total parameters. The company states it achieves 4-12 times greater FLOPs efficiency compared to DeepSeek-V4-Pro-Base, having been trained on 16k GB200 GPUs for 38 days. MAI-Base-2, developed using organic data, is currently undergoing further training before being offered to business clients via Foundry.
Read the full story