Microsoft AI unveils MAI-Base-2, claiming superior performance to OpenAI with a fraction of the compute

Key points

• Microsoft AI CEO Mustafa Suleyman announced MAI-Base-2 during a town hall, stating it surpasses OpenAI’s GPT-5.6 in performance.
• The new model required only one-quarter of the compute power used to train OpenAI’s equivalent model, according to Suleyman.
• MAI-Base-2 features 2.8 trillion total parameters with 80 billion active parameters, utilizing a mixture-of-experts architecture.
• The model was trained on 16,000 Nvidia GB200 GPUs over 38 days using approximately 50 trillion tokens.
• Microsoft claims the model is 4.2 to 12.3 times more compute-efficient than DeepSeek-V4-Pro-Base.
• Suleyman noted that Microsoft has special access to technical information on OpenAI models for these comparisons.

Background

Microsoft has increasingly moved to develop its own AI models to reduce dependence on OpenAI, a key partner and customer. In March 2026, Microsoft reorganized its Copilot products and reassigned Suleyman to focus on AI model development. This move follows a period of strained relations between the two companies, with Microsoft investing billions in OpenAI while simultaneously building its own capabilities. The announcement of MAI-Base-2 is part of this broader strategy to establish Microsoft as a frontier AI developer.

How outlets are covering it

Business Insider highlights the potential conflict of interest, noting that Microsoft’s criticism of OpenAI could impact its $100 billion stake in the company. Investing.com emphasizes the technical specifications and efficiency gains, focusing on the model’s architecture and training process. Moomoo frames the announcement as a potential 'backstabbing' of OpenAI, reflecting the competitive tension between the two firms. Pasquale Pillitteri provides a neutral overview of the model’s specifications, noting that it is not yet available for public testing.

Why it matters

The release of MAI-Base-2 signals a shift in the AI landscape, with major tech companies moving to develop in-house models to reduce reliance on external providers. Microsoft’s claim of superior performance with less compute could have significant implications for the cost and efficiency of AI development, potentially impacting the market dynamics between Microsoft and OpenAI.

What to watch

Microsoft is currently in the post-training phase for MAI-Base-2, with further results expected later. The company has not disclosed commercial availability dates or pricing. Preparations for MAI-Base-3 are already underway, indicating a continued focus on developing its own AI models for products and enterprise customers.