Video test on Wan 2.2 + Fish Speech S2 + DeepSeek models
Overnight, I put it into the machine and let DeepSeek V4.1 Flash generate a nice scenario (of course, based on my voice, which I recorded when I was going to the store, what and how it should look dirty), and it based on this voice
This is mind blowing!!
GPT-6 Astra completed Financial Modeling World Cup challenges four times faster than the winning human competitor using computer use and the people still doing financial modeling manually are not going to enjoy reading that sentence. Not four times faster
After working on DeepSeek v4.1 Flash for dozens of hours, it’s pretty much the same thing.
The method of aligning is basically the same as that of open-pokered, but red has far more dynamic effects and details than red, and the cost of alignment increases significantly.
Imagine kimi slides, but open-source and based on Deepseek Harness
You have reached the end of the archive
All of deepseek