Tilelli: An Open-Source Local LLM Designed to Refuse Instead of Hallucinate

themoroccanship · reddit · 2026-08-06

Tilelli Lab introduced Tilelli, an open-source local LLM aimed at tackling hallucinations. It utilizes a Per-Token Routing architecture with 3 specialized pathways instead of a monolithic design. The model achieves a 0.93 AUROC in catching gibberish and cleanly refuses out-of-distribution inputs. A ternary version is also under active development to bridge the performance gap with standard float models.

Related event: Tilelli: Open-Source LLM Tackles Hallucinations via Per-Token Routing(2 posts)→

Original post →

More from Models

Models channel →