IFM has released the final checkpoint for K2-Horizon-MoVA-36B-A4B, a Mixture-of-Experts model featuring Mixture-of-Values attention that stores 36B parameters but runs only 4B per token. The release includes GGUF formats and links to smaller variants in the K2-Horizon family.
- Achieves frontier-class results on agentic and reasoning benchmarks, outperforming open dense models of ~30B size and MoE models up to 15× its size.
- Supports a native 524,288-token context window from midtraining stages onward.
- IFM will release intermediate checkpoints, training data, recipe, and code to ensure full openness.
The model is positioned as a highly efficient alternative to larger dense models while maintaining competitive performance against closed frontier systems.