# MAD-OPD: Breaking the Ceiling in On-Policy Distillation via Multi-Agent Debate

arXiv: https://arxiv.org/abs/2605.01347 · HTML: https://mlsanity.com/paper/2605.01347/

## Key facts

- Cited by 17 papers in the map (data through 14 Oct 2026).
- Placed in Reinforcement learning for LLM agents, one of 16 directions in LLM agents and tool use.
- Builds on 36 of its 40 references that are in the map.
- First posted on arXiv 2 May 2026.

## Placed in

- LLM agents and tool use / [Reinforcement learning for LLM agents](https://mlsanity.com/agents/reinforcement-learning-for-llm-agents/)
- Knowledge distillation / [On-policy and self-distillation for LLMs](https://mlsanity.com/distill/on-policy-and-self-distillation-for-llms/)

## Builds on (most cited first, 30 of 36)

- [Distilling the Knowledge in a Neural Network](https://mlsanity.com/paper/1503.02531/) (2015)
- [DeepSeekMath: Pushing the Limits of Mathematical Reasoning in Open Language Models](https://mlsanity.com/paper/2402.03300/) (2024)
- [Qwen3 Technical Report](https://mlsanity.com/paper/2505.09388/) (2025)
- [Program Synthesis with Large Language Models](https://mlsanity.com/paper/2108.07732/) (2021)
- [LiveCodeBench: Holistic and Contamination Free Evaluation of Large Language Models for Code](https://mlsanity.com/paper/2403.07974/) (2024)
- [Improving Factuality and Reasoning in Language Models through Multiagent Debate](https://mlsanity.com/paper/2305.14325/) (2023)
- [The Rise and Potential of Large Language Model Based Agents: A Survey](https://mlsanity.com/paper/2309.07864/) (2023)
- [Is Your Code Generated by ChatGPT Really Correct? Rigorous Evaluation of Large Language Models for Code Generation](https://mlsanity.com/paper/2305.01210/) (2023)
- [Encouraging Divergent Thinking in Large Language Models through Multi-Agent Debate](https://mlsanity.com/paper/2305.19118/) (2023)
- [On-Policy Distillation of Language Models: Learning from Self-Generated Mistakes](https://mlsanity.com/paper/2306.13649/) (2023)
- [Sequence-Level Knowledge Distillation](https://mlsanity.com/paper/1606.07947/) (2016)
- [τ2-Bench: Evaluating Conversational Agents in a Dual-Control Environment](https://mlsanity.com/paper/2506.07982/) (2025)
- [Self-Distilled Reasoner: On-Policy Self-Distillation for Large Language Models](https://mlsanity.com/paper/2601.18734/) (2026)
- [Policy Distillation](https://mlsanity.com/paper/1511.06295/) (2015)
- [Reinforcement Learning via Self-Distillation](https://mlsanity.com/paper/2601.20802/) (2026)
- [ReConcile: Round-Table Conference Improves Reasoning via Consensus among Diverse LLMs](https://mlsanity.com/paper/2309.13007/) (2023)
- [OpenThoughts: Data Recipes for Reasoning Models](https://mlsanity.com/paper/2506.04178/) (2025)
- [Rethinking On-Policy Distillation of Large Language Models: Phenomenology, Mechanism, and Recipe](https://mlsanity.com/paper/2604.13016/) (2026)
- [ToolACE: Winning the Points of LLM Function Calling](https://mlsanity.com/paper/2409.00920/) (2024)
- [On-Policy Context Distillation for Language Models](https://mlsanity.com/paper/2602.12275/) (2026)
- [Learning beyond Teacher: Generalized On-Policy Distillation with Reward Extrapolation](https://mlsanity.com/paper/2602.12125/) (2026)
- [Entropy-Aware On-Policy Distillation of Language Models](https://mlsanity.com/paper/2603.07079/) (2026)
- [A Survey of On-Policy Distillation for Large Language Models](https://mlsanity.com/paper/2604.00626/) (2026)
- [Revisiting On-Policy Distillation: Empirical Failure Modes and Simple Fixes](https://mlsanity.com/paper/2603.25562/) (2026)
- [Self-Distilled RLVR](https://mlsanity.com/paper/2604.03128/) (2026)
- [Privileged Information Distillation for Language Models](https://mlsanity.com/paper/2602.04942/) (2026)
- [Debate or Vote: Which Yields Better Decisions in Multi-Agent Large Language Models?](https://mlsanity.com/paper/2508.17536/) (2025)
- [Chain-of-Agents: End-to-End Agent Foundation Models via Multi-Agent Distillation and Agentic RL](https://mlsanity.com/paper/2508.13167/) (2025)
- [VitaBench: Benchmarking LLM Agents with Versatile Interactive Tasks in Real-world Applications](https://mlsanity.com/paper/2509.26490/) (2025)
- [Stable On-Policy Distillation through Adaptive Target Reformulation](https://mlsanity.com/paper/2601.07155/) (2026)

## Built on by (most cited first, 17 of 17)

- [Tailoring Teaching to Aptitude: Direction-Adaptive Self-Distillation for LLM Reasoning](https://mlsanity.com/paper/2605.22263/) (2026)
- [PCSD: Persistent Consistency for Self-Distillation in Agentic Reinforcement Learning](https://mlsanity.com/paper/2608.01837/) (2026)
- [A Formula-Driven Survey and Research Agenda for On-Policy Distillation](https://mlsanity.com/paper/2606.22793/) (2026)
- [From Proprietary to Open-Source: Bridging the Distribution Gap via Multi-Agent Protocol Distillation in Agentic Search](https://mlsanity.com/paper/2607.24280/) (2026)
- [When Top-K Misses the Decision: Tool-Call Drift in Multi-Teacher On-Policy Distillation](https://mlsanity.com/paper/2607.07050/) (2026)
- [The Physics of Multi-Turn Long-Horizon Planning: From Pre-training to Post-training via Single- and Multi-Teacher On-Policy Agentic Distillation](https://mlsanity.com/paper/2607.24720/) (2026)
- [Distill What You Trust: Reliability-Aware Multi-Teacher On-Policy Distillation](https://mlsanity.com/paper/2609.23697/) (2026)
- [MIRROR: Learning from the Other View for Multi-Modal Reasoning](https://mlsanity.com/paper/2607.21552/) (2026)
- [From Parameters to Behaviors: A Survey of Model Fusion for Large Language Models](https://mlsanity.com/paper/2609.19553/) (2026)
- [Uncertainty Quantification for LLM Agents: A Taxonomy, an Evaluation Protocol, and an Empirical Study](https://mlsanity.com/paper/2609.07395/) (2026)
- [From Self-Distillation to Self-Practice: Privileged Information for Multi-Turn Agents](https://mlsanity.com/paper/2609.29051/) (2026)
- [PUBG Ally: A Conversational Embodied Agent as an AI Teammate](https://mlsanity.com/paper/2609.29837/) (2026)
- [ChemOPD: Multi-Teacher On-Policy Distillation for Multi-Task Chemical Reasoning](https://mlsanity.com/paper/2609.33838/) (2026)
- [MAS-OPD: On-Policy Distillation for Multi-agent Systems](https://mlsanity.com/paper/2609.34234/) (2026)
- [GraphOPD: Graph-Augmented On-Policy Distillation for LLM Agents](https://mlsanity.com/paper/2610.08959/) (2026)
- [State-OPD: State-Grounded On-Policy Distillation for Multi-Turn Agents](https://mlsanity.com/paper/2610.13238/) (2026)
- [Heterogeneous Agentic Coevolution](https://mlsanity.com/paper/2610.16078/) (2026)

## Abstract (arXiv)

On-policy distillation (OPD) trains a student on its own trajectories under token-level teacher supervision, but existing methods are capped by a single-teacher capability ceiling: when the teacher errs, the student inherits the error. OPD also remains largely unexplored in agentic tasks, where per-step errors compound across long trajectories and destabilize training. We propose MAD-OPD (Multi-Agent Debate-driven On-Policy Distillation), which breaks this ceiling by recasting the distillation teacher as a deliberative collective of teachers that debate over the student's on-policy state; the debate produces an emergent collective intelligence that supplies token-level supervision, with each teacher's contribution weighted by its post-debate confidence. To extend OPD to agentic tasks, we also introduce On-Policy Agentic Distillation (OPAD), which adds step-level sampling to stabilize training under multi-step error compounding. We additionally derive a task-adaptive divergence principle, selecting JSD (Jensen-Shannon divergence) for agentic stability and reverse KL (Kullback-Leibler) divergence for code generation, and verify it both theoretically and empirically. Across six teacher-student configurations (Qwen3 and Qwen3.5; 1.7B-14B students, 8B-32B teachers) and five agentic and code benchmarks, MAD-OPD ranks first across all six configurations; on the 14B+8B→4B setting it lifts the agentic average by +2.4% and the code average by +3.7% over the stronger single-teacher OPD.