Search
China’s Open-Source Marvel: Kimi K2 Wins Praise from Vercel CEO, Besting International Closed-Source Models in Agent Use Cases
2025-10-22 /
Read about 0 minute
Author：小编
As reported by 36Kr, Guillermo Rauch, the CEO of the American startup Vercel, took to social media to share that internal agent testing frameworks had been employed to assess the real-world task performance of multiple mainstream large language models. The findings were striking: Kimi K2, an open-source model crafted by China’s Moonshot team, demonstrated a remarkable fivefold increase in speed and a 50% boost in accuracy compared to GPT-5 when deployed in agent application scenarios. This stellar performance not only eclipses that of GPT-5 but also outshines other internationally renowned closed-source models, such as Claude Sonnet4.5, sparking widespread acclaim within the tech community.
-
C114 Communication Network -
Communication Home
7 X 24 Track global technological trends
