K2 Horizon MoVA 36B A4B

K2 Horizon MoVA 36B-A4B is the sparse mid-size member of IFM's (MBZUAI) fully open K2 Horizon family: a Mixture-of-Experts reasoning model with 36B total and about 4B active parameters per token (48 layers, 100 routed experts with 8 active plus 1 shared expert) that also uses IFM's new Mixture-of-Value Attention (MoVA), where a router selects a few value-projection experts per token. Pretrained on 22.9T tokens with a native 524,288-token (512K) context window from midtraining onward. Text in, text out, with per-request reasoning effort (high recommended) and JSON/XML tool-call formats. Released 2026-09-03 under Apache 2.0 with BF16 weights plus FP8 and GGUF builds, intermediate checkpoints, training data, code, and logs. No priced provider is cataloged yet.