MiMo-V2.6: Scaling Reinforcement Learning Towards Self-Improvement

What happened: A technical report introduces MiMo-V2.6, an omni-modal model family trained with scaled reinforcement learning, with mid-training on a broad multimodal corpus beforehand. Key numbers: 1,568 samples and 2.7-3.7B tokens per step, context lengths up to 1M. Why it matters: The report states the work open-sources training dynamics, RL environments, and the RL framework to support reproduction and further research on scaled RL and model self-improvement.

A technical report introduces MiMo-V2.6, an omni-modal model family trained with scaled reinforcement learning, with mid-training on a broad multimodal corpus beforehand.

The report states the work open-sources training dynamics, RL environments, and the RL framework to support reproduction and further research on scaled RL and model self-improvement.