Open-source decision model Laya hits 35k stars, outputs calibrated probabilities in 33ms

kingslyj · x · 2026-11-20

Kerala-based researcher Nandakishor open-sourced Laya, a "decision model" that skips text generation: one forward pass (33ms) yields calibrated probabilities for tasks like ticket routing, jailbreak detection, and urgency scoring—replacing slow, costly LLM+JSON calls. Born from two arXiv papers (Mar 2025, Oct 2025), it was released under Apache 2.0 four days after a funded lab shipped the same concept as a closed paid API. Results: 35,683 stars and 3,200 forks in two months, 1,363 points on HN, #1 trending on Hugging Face with 374 community models, and 250k monthly PyPI downloads.

Original post →

More from Models

Models channel →