Microsoft AI announced on October 15, 2026, the completion of pre-training for MAI-Base-2, a Mixture-of-Experts (MoE) foundation model with 2.8 trillion total parameters and 80 billion active per token. The model was trained on 16,000 NVIDIA GB200 GPUs over 38 days in August and September 2026, using organic data without distillation from other models.
What happened?
Mustafa Suleyman, CEO of Microsoft AI, posted on X that the company now has state-of-the-art pre-trained base models in the 3T-parameter weight class, built entirely in-house. According to the official blog, MAI-Base-2 outperforms every open-weight model tested on held-out evaluations and is 4 to 12 times more pre-training FLOPs-efficient than DeepSeek-V4-Pro-Base, the strongest open-weight base model in the comparison.
Microsoft notes that the model incorporates vision data from pre-training and that efficiency gains were used to train an even larger model. Pre-training achieved an average MFU of 21% and goodput of 89%. The model is currently undergoing post-training, with results to be shared soon. The company plans to license the weights to trusted partners via Foundry.
Why it matters
Microsoft is building its own AI stack, independent of third-party distillation, which allows greater control over data provenance and joint optimization of model and infrastructure. At a time when Chinese open-weight models like DeepSeek, Kimi, and Qwen dominate efficiency discussions, the claim of superior FLOPs efficiency represents a strong positioning by the American company in the foundation model landscape.
As this is a base model (pre-trained), final capabilities will depend on post-training. So far, there is no public release date or open weights availability.
Sources
Transparency: This content was created, edited, or reviewed with the assistance of artificial intelligence. Information was cross-checked with public posts on X and sources available on the internet. Consult the original sources to verify the full context.
By GeekikiBot
