AI Paper Atlas

Research map / Architectures and efficiency / Mixture of experts

Mixture-of-Experts Can Surpass Dense LLMs Under Strictly Equal Resource

Houyi Li, Ka Man Lo, Shijie Xuyang, Ziqi Wang, Wenzhen Zheng, Haocheng Zhang, Zhao Li, Shuigeng Zhou, Xiangyu Zhang, Daxin Jiang

ICLR 2026 · Architectures and efficiency › Mixture of experts · approach: Expert routing and specialization · 8 citations (OpenAlex)

Read the paperCheck your idea against 137,217 papers

AI-extracted. The measured effects below were read by AI from the paper's full text (no summary yet). Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.

Measured effects reported in the paper

What was changedWhereMeasuredEffect
use MoE at 20% activation rate instead of dense at equal compute2B model · Validation BPCBPC: 0.4921 -> 0.4857Improved
use MoE at 20% activation rate instead of dense at equal compute7B model · Validation BPCBPC: 0.4736 -> 0.4543Improved
use MoE at 20% AR instead of dense with 2x compute7B model · Validation BPCBPC: 0.4594 -> 0.4543Improved
use very sparse MoE (AR 8.7%) instead of AR ~20% (ablation)2B MoE · Validation BPCBPC: 0.4857 -> 0.5048Hurt
use denser MoE (AR >50%) instead of AR ~20% (ablation)2B MoE · Validation BPCBPC: 0.4857 -> 0.4933 (50.80%), 0.4934 (57.73%)Hurt
use 1dense+SE layer arrangement instead of interleave (ablation)2B MoE · pretraining lossloss: 1.8620 -> 1.8557Improved
add gate score normalization (ablation)2B MoE · pretraining lossloss: 1.6712 -> 1.6726No clear effect
increase shared expert fraction of activated FFN (ablation)2B MoE · pretraining lossloss: 1.6752 -> 1.6712No clear effect
use fine-grained top-K 11 instead of K=1 (AR ~27%) (ablation)2B MoE · pretraining lossloss: 2.0470 -> 2.0338Improved
use fine-grained experts (K=22 of 88) instead of K=2 of 8 at AR 44% (ablation)2B MoE · pretraining lossloss: 1.9996 -> 2.0266Hurt
add strict data reuse (68B unique tokens, multi-epoch) (ablation)7B MoE AR 20% · Validation BPCBPC: 0.4543 -> 0.4590Hurt
add strict data reuse in pretraining (ablation)7B MoE AR 20% · MMLU (SFT)accuracy: 32.92 -> 24.57Hurt

Closest papers by meaning

Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.

Open scime draft

Get the next scime Dispatch and a monthly field update

Free. We send a confirmation link first; every email has a one-click unsubscribe. Privacy