On Kimi K3: Its Capabilities And Related Discontents

Moonshot AI recently released Kimi K3, a large language model with impressive benchmarks that would become the strongest open model if weights are released as planned. But LessWrong cautions against reading too much into the hype: the model still lags frontier closed models by four to six months, and its benchmark scores benefit from maximum-effort testing using significantly more tokens than typical comparisons. At two point eight trillion parameters—the largest open model released so far—Kimi K3 excels at coding and three-D tasks but is slow and computationally expensive. As Bloomberg reports, Moonshot plans a Hong Kong IPO within the next six months.

Listen to this story

Hear this and more stories in a personalized audio briefing.

Open The Chonkerton