GLM 5.1 and DeepSeek 4 are acceptable, but the cost of hardware and energy cost that depending on your use case you may as well purchase a Tokens. They get useless and stupid rapidilty if you quant enough to run on single 16-24GB GPU style.