Breaking the Reward Barrier: Accelerating Tree-of-Thought Reasoning via Speculative Exploration cs.LG · 2026-05-11