Community curated by people like you

Olmo stories

Oct 2, 2026

Ai2 releases OlmoCore 3 training framework for mixture-of-experts models

Allen Institute for AI released OlmoCore 3, an open development framework for large mixture-of-experts language models that delivers 2.7x throughput gains.

Oct 2, 2026·2 sources

00

Top claims

▪Publicly documenting unresolved training failures is necessary for scientific progress in AI
▪OlmoCore 3 reached a peak throughput of 858 TFLOPS per GPU during a 1.2-trillion-parameter configuration benchmark using 512 NVIDIA B300 GPUs under artificial random routing
▪The Allen Institute for AI documented 'token gerrymandering,' a failure mode where the auxiliary loss metric for routing balance improves while the actual distribution of work across experts degrades

Topics

Mixture of experts (MoE)Allen Institute for AI Open-source AI AI foundation models Open model deployment tools Compute, chips & AI infrastructure