I benchmarked IFM/K2-Horizon-7B on 16GB VRAM

A user benchmarked IFM/K2-Horizon-7B on 16GB VRAM, finding it works but performs significantly worse than Qwen-3.8-27B. The Qwen3.8-27B-GSQ-RCO-IQ3_XXS-mtp model achieved 15/15 completed tasks with 1.000 correctness in 348.5 seconds. In contrast, K2-Horizon-7B-Q8_0 completed 11/15 tasks with 0.909 correctness in 498.1 seconds. Ornith-1.5-9B-Q8_0 and Qwen3.5-9B-Q8_0.gguf (MTP) also showed strong performance.