Today for AI
HOT RADAR
llmHEAT 11.9°

AI CURATED HOT STORY · 10/16/2026

Microsoft Launches MAI-Base-2: Claims Performance Surpasses GPT-5.6 Underlying Model with One-Quarter Compute

2 reports archived1 independent sourcesupdated 10/16/2026, 11:34:25
Synthesis & Latest Updates
1 Sources Cross-Validated

Microsoft AI CEO Mustafa Suleyman announced that the in-house foundation model MAI-Base-2 outperforms the OpenAI internal model powering GPT-5.6 during pre-training, using only one-quarter of the compute. The model also claims superior computational efficiency compared to DeepSeek, requiring just one-twelfth of the compute, signaling Microsoft's strategic move toward an independent and efficient LLM infrastructure.

LATEST/Microsoft AI CEO Mustafa Suleyman announced that the in-house foundation model MAI-Base-2 outperforms the OpenAI internal model powering GPT-5.6 during pre-training, using only one-quarter of the compute. The model also claims superior computational efficiency compared to DeepSeek, requiring just one-twelfth of the compute, signaling Microsoft's strategic move toward an independent and efficient LLM infrastructure.

HEAT TRENDHourly heat curve

2 fully observed hours
Fewer than 3 fully observed hours; trend not plotted yet.

TIMELINECoverage timeline

Total 2 reports · Latest first
  1. IT之家 · 智能时代T2·86 pts
    • MAI-Base-2 features 2.8T total parameters and 80B active parameters, leveraging MoE architecture for high performance at lower inference costs.
    • Claims 4x to 12x higher pre-training compute efficiency compared to DeepSeek-V4-Pro-Base, highlighting significant training optimization advantages.
    • Success attributed to scientific scaling, model-infrastructure co-design, and data scaling, with development of MAI-Base-3 already initiated.
  2. IT之家 · 智能时代T2·78 pts
    • MAI-Base-2 outperforms the internal OpenAI model (codenamed Spud) powering GPT-5.6 in pre-training tasks.
    • Microsoft claims the model requires only 1/4 of the compute used by comparable OpenAI models, significantly reducing infrastructure costs.
    • Compared to DeepSeek, Microsoft asserts higher computational efficiency, needing just 1/12 of the compute for equivalent or better results.