AI Briefing
KO

AI2 unveils EMO

·2026.05.09 05:57

Key point

AI2 has unveiled EMO, a 1B active/14B total MoE model.

Details

AI2 has unveiled a new MoE model called EMO.

  • Its scale is 1B active / 14B total.
  • It was trained on 1T tokens.
  • The key feature is document-level routing, where experts cluster by domain—such as health and news—rather than by surface patterns.
  • The model collection is available on Hugging Face: https://huggingface.co/collections/allenai/emo
mixture-of-expertsai2emodocument-routingsparse-activation
Read originalShare on Threads

This summary was generated automatically by AI. Check the original for the author's claims and context. Copyright belongs to the original author.

Our guide explains how the AI works. Report summary errors, attribution issues, or removal requests via Contact.

Related

HuggingFace BlogAI2 Releases Benchmark Analysis MethodHuggingFace BlogQwen3.8 2.4T ReleasedNVIDIA BlogNVIDIA Nemotron 3.5 Lightning and NeMo Switchyard Deliver Faster, Smarter, and More Efficient Agentic AIPyTorchKRAgents-A1 Released, Achieving 1T-Class Performance with a 35B ModelNVIDIA BlogWhy 'Performance per Watt' Is the Ultimate Metric for AI Infrastructure EfficiencyHacker NewsLongCat-2.0, a large-scale MoE model with 1.6T total parameters and 48B active parameters