Kimi K3, and what we can still learn from the pelican benchmark

Moonshot AI released Kimi K3, a new large language model that achieves state-of-the-art results on the Pelican benchmark, outperforming GPT-4o and Claude 3.5 Sonnet. The Pelican benchmark, designed to test long-context understanding and reasoning, reveals that even top models struggle with tasks requiring deep integration of information across extended passages, highlighting ongoing challenges in AI development.