Search
Guillermo, the CEO of the U.S.-based startup Vercel, recently shared in a post that during internal real-world benchmark tests involving agent-based model series, the Chinese open-source model, Kimi K2, exhibited remarkable performance. When compared to leading closed-source models such as GPT-5 and Claude Sonnet 4.5, Kimi K2 proved to be five times faster and 50% more accurate. Additionally, it can seamlessly switch models at no extra cost, thanks to its advanced operational platform. According to data released by Guillermo, for identical tasks, GPT-5 and Sonnet 4.5 require 8-10 minutes to complete, with accuracy rates falling below 40% and 50%, respectively. In stark contrast, Kimi K2 finishes the same tasks in just 2 minutes and achieves an accuracy rate exceeding 60%, outperforming both in terms of both speed and precision.
-
C114 Communication Network -
Communication Home
