The K2 Horizon 7B model ranks between Qwen 3.6 27B and Qwen 3.6 35B on the Artificial Analysis Intelligence Index. Early testing shows it can compile the latest llama.cpp for CUDA successfully, suggesting strong performance relative to its size. If it holds to this score, it offers surprisingly capable inference for GPU‑poor setups.