News for hackers. Stuff that matters.
Re:Arcee Trinity Mini: US-Trained Moe Model by htrp on Monday December 01, 2025 @08:49PM (#46116350)
Trinity Nano Preview: 6B parameter MoE (1B active, ~800M non-embedding), 56 layers, 128 experts with 8 active per token

Trinity Mini: 26B parameter MoE (3B active), fully post-trained reasoning model

They did pretraining on their own and are still training the large version on 2048 B300 GPUs

There are no comments on this comment yet.