Piccolo: Recursive Self-Improvement via Agentic Data Engineering
November 17, 2026
Piccolo is a family of multimodal decision models from 0.8B to 27B parameters that use recursive self-improvement to coevolve model capabilities and training data. The Piccolo-27B model achieved a score of 75.59 on the Decision Index Vision Full leaderboard.
HOW THIS AFFECTS YOU
●
builderYou can deploy smaller, highly capable decision models that have been optimized through self-improvement loops.
●
researcherYou can use agentic workflows to automatically generate and refine training data for model improvement.
●
founderThis shows a path toward scaling model intelligence through automated data engineering rather than just more human data.