STILL (Slow Thinking with LLMs)

RUCAIBox's (Wayne Xin Zhao) slow-thinking series — the earliest open o1/R1 reproduction reports (767★): STILL-1 (Nov 2024) reward-guided tree search, STILL-2 (Dec 2024) imitate-explore-self-improve distillation, STILL-3 (Mar 2025) R1-style RL yielding the STILL-3-1.5B and TOOL-32B models. BAAI and DataCanvas provided compute only; the reports predate most open R1 replications.