I saw on Xiaohongshu that someone used openminis to install deepseek harness... It’s really hard work 😂
I tried it with DeepSeek V4 Flash (max), but maybe the question is too difficult...
On the contrary, it may be good to see superiority and inferiority.
DeepSeek V4 Flash needs 2× RTX 6000s
Qwen3.8-27B FP8 needs just 2× RTX 4090s (2,873 tok/s prefill, 74 tok/s decode) I was curious how they compared, so I ran the same Pi agentic task on both. Both solved it in one shot with no follow-ups Here are the outputs side by side👇
What makes me feel very interesting is the “track” behind DeepSeek Harness
You can clearly see the thinking behind the model How raw input and output becomes user UI step by step And you can view the conversation details and Token consumption in chronological order This reminds me of Chrome’s Dev tool The idea that everything is a plug-in also reminds me of Chrome…
You have reached the end of the archive
All of deepseek