M3OPD-Reasoner opens multimodal post-training recipe
The M3OPD-Reasoner paper presents an open recipe for post-training vision-language models across image STEM, image perception, and video reasoning. Starting from Qwen3.5-9B, the authors report that multi-teacher on-policy distillation produced a 9B student closely matching each teacher in its domain and outperforming mixed-training baselines on average across 21 benchmarks; they also released the datasets, training code, and stage checkpoints.
The paper offers practitioners a fully open, reproducible workflow that combines domain-specific teachers, on-policy distillation, and subsequent reinforcement learning for multimodal reasoning, though its reported results are based on the authors’ evaluation.
Computer Science > Computer Vision and Pattern Recognition
arXiv:2611.23034 (cs)
[Submitted on 17 Nov 2026]
Title:M3OPD-Reasoner: An Open Recipe for Post-Training Multimodal Reasoning with Multi-Teacher On-Policy Distillation
Authors:Nguyen Quang Trung, Hoang Phan, Shuo Sun, Longyin Zhang, AiTi Aw
View a PDF of the paper titled M3OPD-Reasoner: An Open Recipe for Post-Training Multimodal Reasoning with Multi-Teacher On-Policy Distillation, by Nguyen Quang Trung and 4 other authors
#M3OPD-Reasoner-9B#Qwen3.5-9B#Multi-Teacher On-Policy Distillation#Group Sequence Policy Optimization#M3OPD-SFT-1.35M#M3OPD-RL-193K