DeepSeek V5 is expected in September, and their performance claims have become outrageous.
Rumors state that DeepSeek is making V5 on an entirely new platform and not the old one of V4, which has better reasoning, coding, and agents.
However, the performance claim that needs verification is that of a Mythos-level performance despite being open-weight and significantly less costly.
It is far more impressive than outperforming V4.
As there is no benchmark for V5 publicly available, I do not claim it to be a Mythos-killer.
In case DeepSeek manages to get close enough to Mythos but remains an open-weight system, the entire open-weight world will change overnight.
Would you use V5 over Mythos in case it achieves 90-95% of Mythos' performance at a reduced price?
ENPIRE -> ASPIRE, our 2nd work in the series for Physical AutoResearch. We are building the components for robot self-improvement, one /skill at a time.
When AI makes implementation cheap, abstract debates become expensive.
Teams should stop arguing in slides and start validating ideas with prototypes and code.