IFM/K2-Horizon-MoVA-36B-A4B

IFM releases K2-Horizon-MoVA-36B-A4B, a sparse Mixture-of-Experts model with 36 billion parameters but only 4 billion active per token. The card provides vLLM serving instructions and benchmark comparisons against other open-weight models. Training code and data remain pending, so independent verification of the reported efficiency gains is not yet possible.