Microsoft said it completed pre-training MAI-Base-2, a 2.8-trillion-parameter foundation model that it says outperformed the open-weight models it evaluated.
Microsoft AI said on Oct. 15, 2026, that it had completed pre-training MAI-Base-2, a mixture-of-experts foundation model with 2.8 trillion total parameters and 80 billion active per token. The 38-day training run used 16,000 Nvidia GB200 GPUs and roughly 50 trillion tokens; post-training is still underway. Microsoft says the model beat every open-weight system it tested on held-out evaluations, while estimated pre-training compute efficiency was 4.2–12.3 times that of DeepSeek-V4-Pro-Base across four tests. Those figures are company-reported comparisons, not evidence of universal superiority: tests used bits per byte, where lower is better, and some competitors were post-trained checkpoints. The result matters because a more compute-efficient base model could lower the cost of building future systems and strengthen Microsoft’s in-house AI capability amid competition with DeepSeek and other open-model developers.