Parcae
Doing more with fewer parameters using stable looped models
Parcae is a stable looped language-model architecture/family developed with UCSD/Sandy Research and Together AI collaboration. It reuses layers through a recurrent block to scale compute by recurrence rather than parameter count; released pretrained base models include 140M, 370M, 770M, and 1.3B parameter variants.
Pricing
No public numeric price for Parcae was found on Together AI’s official pricing or serverless model pages; treat as not publicly listed.
Together AI’s official Parcae blog post confirms the model exists, but the current public Together AI pricing page and serverless model pricing docs do not list a Parcae entry or any numeric rate. As of 2026-04-16, no public official pricing could be found for Parcae.