The user runs Qwen 3.6 35B on a modest laptop (16 GB DDR5, 13th‑gen i5, no GPU) via llama.cpp at ~15 output tokens per second. They desire a smarter Mixture‑of‑Experts model that is larger than Qwen 3.6 35B but smaller than GLM 4.5. However, they have found no publicly available models that fit this intermediate size range.