Microsoft AI unveils 2.8 trillion-parameter model, claims efficiency gains

Investing.com -- Microsoft has completed pre-training a new AI model with 2.8 trillion total parameters, saying it outperformed every open-weight model it tested while requiring up to 12 times less computing power to match a leading rival.

The model, MAI-Base-2, has 80 billion active parameters and uses a mixture-of-experts architecture that activates only a portion of its parameters for each task. Microsoft said its training approach was 4.2 to 12.3 times more compute-efficient than that of DeepSeek-V4-Pro-Base across four evaluations.

The announcement marks another step in Microsoft’s push to develop its own frontier AI models and reduce its reliance on external providers. The company said Thursday that MAI-Base-2 is undergoing post-training, with further results expected later.

Microsoft trained the model on 16,000 Nvidia GB200 GPUs over 38 days in August and September, using a corpus of roughly 50 trillion training tokens. Its dataset of unique tokens grew about 36% from the previous model, MAI-Base-1, released in June.

The company said it trained the model on original data rather than distilling knowledge from other AI systems, a distinction it sees as important for licensing its model weights to trusted partners through Azure AI Foundry.

Microsoft attributed the performance gains to improvements in data quality, model architecture and training infrastructure. It also expanded training to include vision natively, with image-related material accounting for 18% of the training mixture.

The company said its training recipe was 6.3 to 12.7 times more compute-efficient than MAI-Base-1, reflecting advances in optimization and model design.

Microsoft did not disclose commercial availability dates or pricing. It said preparations for MAI-Base-3 are already underway as it continues developing its own AI models for its products and enterprise customers.

Original Article

You May Also Be Interested In

• Microsoft (MSFT) PT Raised to $675 at UBS as 'Valuation Multiple Is Not Stretched'
• International Paper rises on report of private equity interest in European arm
• Morgan Stanley upgrades Quantinuum to Overweight on technology edge; shares rise