Our accuracy-first reasoning model, which works through hard mathematics and code at length and reaches its base model's accuracy with far fewer tokens.

Our efficiency-first reasoning model, which answers briefly and almost never falls into a loop, made for phones and everyday questions.

Our compact open model, for instructions, multi-turn work, code, and efficient reasoning.

Paged mixture-of-experts packages that run models larger than device memory.

GGUF · Expert paging

Public-domain reference text for Noema's on-device retrieval.

Dataset · Public domain