MiMo-V2.6-Pro-RL

MiMo-V2.6-Pro-RL is the flagship checkpoint of the MiMo-V2.6 series developed by the Xiaomi MiMo team. It is an omnimodal model with a sparse Mixture-of-Experts architecture totaling 1.02 trillion parameters, of which 42 billion are activated per forward pass (384 routed experts, 8 active). The model accepts text, image, video and audio, generates text, and supports a context window of up to 1 million tokens.

Architecture

The LLM backbone has 70 transformer layers (60 sliding-window attention and 10 global attention) with a hidden size of 6144. It combines omnimodal encoders: a MiMo ViT vision encoder (681M parameters, 28 layers) and audio encoders (308M tokenizer + 127M patch encoder). Speculative decoding is handled by a 5-layer Multi-Token Prediction drafter. Weights are released in FP8 format.

How the Pro-RL variant differs

Pro-RL is the largest, flagship variant of the series (alongside the smaller MiMo-V2.6-Flash-RL) and the centerpiece of the "scaling reinforcement learning toward self-improvement" approach. It undergoes a single mixed RL run spanning coding, general agents, visual tasks and cybersecurity, using fully asynchronous GRPO on large batches (1,568 prompts × 16 rollouts per step). The reward signal is scaled by groupwise agentic grading (Groupwise Reward Synthesis and Groupwise Advantage Redistribution), and after RL the model is refined with multi-prefix multi-teacher on-policy distillation (MOPD2).