Microsoft finishes pre-training MAI-Base-2: 2.8T MoE, 4-12x more FLOPs-efficient than DeepSeek

gordic_aleksa · x · 2026-10-16

Microsoft AI completed pre-training its new flagship base model, per the official "Frontier Pre-training at Microsoft AI" blog:

• MAI-Base-2: an MoE foundation model with 80B active / 2.8T total parameters
• Outperforms every open-weight model tested on Microsoft's held-out evals
• 4-12x more pre-training FLOPs-efficient than the strongest open base model, DeepSeek-V4-Pro-Base, with savings reinvested into a larger model
• Follows June's MAI-Thinking-1, post-trained on MAI-Base-1
• The team also shares scaling-law insights, including dropout scaling laws

A major milestone for Microsoft's in-house frontier-model effort.

Related event: Microsoft AI Unveils MAI-Base-2, a 2.8T-Parameter MoE Foundation Model with Completed Pre-training(5 posts)→

Original post →

More from Models

• EpochAI pegs GPT-6 training at up to 1.8e27 FLOP; scaling01 calls param estimates way too high — scaling01 · 2026-10-16
• Gemini 4 Argon reportedly now available to all Antigravity Pro users — clmt · 2026-10-16
• Mythos on Claude's 'smallness': accurate proprioception, not low self-esteem — repligate · 2026-10-16
• Meta's Muse Spark Open scores 39 vs 48 for closed Spark 1.3 on AA Intelligence Index — EAccelerate_42 · 2026-10-16
• PhoneLLM Mini (30B) and Max (120B) target sub-100ms voice AI latency on Workers AI — ritakozlov · 2026-10-16
• Yacine on LLMs: "I'm sick of how stupid these language models are. Accelerate!" — yacineMTB · 2026-10-16

Models channel →