alphaXiv(@askalphaxiv):A new post-training paradigm is emerging in 2026: Multi-Teacher On-Policy Distillation (MOPD), which lets a single model absorb capabilities from multiple specialized RL teachers. It’s already being used in frontier-scale models like MiMo-V2-Flash, Kimi K3, and DeepSeek-V4, making MOPD an increasingly important technique for combining specialized capabilities into one model. We have compiled a list of popular papers tracing the development of MOPD from its earliest forms to its latest iterations. Check it out now:

2026.09.02 20:50

A new post-training paradigm is emerging in 2026: Multi-Teacher On-Policy Distillation (MOPD), which lets a single model absorb capabilities from multiple specialized RL teachers. It’s already being used in frontier-scale models like MiMo-V2-Flash, Kimi K3, and DeepSeek-V4, making MOPD an increasingly important technique for combining specialized capabilities into one model. We have compiled a list of popular papers tracing the development of MOPD from its earliest forms to its latest iterations. Check it out now:

显示更多