What is the best intelligence/stable model currently for a single GB10/DGX spark?

Is Qwen 3.6 27b still the go' ol' reliable at this point? I know 35b is faster but it just doesn't give as good results. Is it possible to run deepseek v4 flash on a single spark at decent tk/s without having to ssd stream or Q1 lobotomize? I was having good hopes for laguna s 2.1 but so far ive see

中文全文翻译正在补齐，可先查看下方英文原文。

Is Qwen 3.6 27b still the go' ol' reliable at this point? I know 35b is faster but it just doesn't give as good results. Is it possible to run deepseek v4 flash on a single spark at decent tk/s without having to ssd stream or Q1 lobotomize? I was having good hopes for laguna s 2.1 but so far ive see