Equilibrium Reasoners: Learning Attractors Enables Scalable Reasoning

Equilibrium Reasoners learn task-conditioned attractors, enabling scalable latent reasoning through deeper iterations and broader stochastic trajectories at test time.