K2 Horizon 7B

K2 Horizon 7B is a 7-billion-parameter dense language model developed by the Institute of Foundation Models (IFM) at the Mohamed bin Zayed University of Artificial Intelligence (MBZUAI). Released in September 2026 as part of a six-model family, it is designed for on-device applications and efficient single-GPU serving. The model is released under an Apache 2.0 license and is notable for its "fully open" approach, which includes the publication of weights, training code, data recipes, and intermediate checkpoints to ensure full reproducibility.

Technically, the model is a decoder-only architecture that features a native context window of 524,288 tokens (512K). This long-context capability was integrated during the mid-training stages, allowing the model to handle massive codebases and extensive document sets. K2 Horizon 7B is specifically positioned as the threshold size within its model fleet where complex software engineering tasks and multi-step agentic workflows become practical.

Performance benchmarks highlight the model's strengths in coding and agentic reasoning. It achieved a score of 70.6 on SWE-bench Verified and 39.1 on Terminal-Bench 2.1, the latter measuring its ability to execute tasks within a Linux shell environment. The development process follows the LLM360 transparency framework, providing researchers with a comprehensive record of the training lifecycle, including the pre-training corpus and fine-grained logs.