Microsoft AI has completed the pre-training phase for its new foundation model, MAI-Base-2, a mixture-of-experts network with 2.8 trillion total parameters. The training process, which utilized 16,000 Nvidia GB200 GPUs over 38 days, demonstrated significant FLOPs efficiency compared to other models. Microsoft plans to license the model weights to trusted partners through its Foundry platform after post-training, with no open-weight release or specific launch details announced. AI
IMPACT Sets a new benchmark for large-scale model training efficiency and signals a shift towards licensed access for advanced foundation models.
RANK_REASON Frontier-lab model release with system card. [lever_c_demoted from frontier_release: ic=1 ai=1.0]
AI-generated summary · Google Gemini · from 1 sources. How we write summaries →