Tp4 didn't use a switch (saves money!).
DeepSeek-V4.1-Flash on 4 DGX will continue to be my main model going forward. Thanks to mialab.
Given that WOW is in the news, Deepseek 4.1 flash (left) vs GLM 5.3 Flash (right) making Jimothy's…
Given that WOW is in the news, Deepseek 4.1 flash (left) vs GLM 5.3 Flash (right) making Jimothy's penthouse in "minecraft" (both quantized to fit on 2 sparks). Deepseek looks better. I couldn't even zoom out enough on GLM to get a better view. Setup (same for both) Same…
By popular demand we have added deepseek_ai and googlegemma models!!
You can now compare Kimi K3, Deepseek V4 Pro and Gemma 4 31B models against your real knowledge work for free! what models should we add next? shoutout MalayVasa for the interface makeover, absolutely
Founder & CEO of says they recently switched one process from Gemini to DeepSeek for better performance, lower cost & more predictable results.
Their approach: constantly test AI models and use whatever works best.
You have reached the end of the archive
All of deepseek