Kimi K3: Performance and the Controversies That Come With It (70-minute read)

Kimi K3 is a 2.8 trillion parameter model, currently the largest open-weight model, and it shows the strongest performance among open models on benchmarks. However, compared to the closed-model frontier, it lags by at least 4 months, roughly 6 months at the median, with the pre-training gap being particularly large.

Benchmark performance was measured under conditions using the maximum number of tokens, which may overstate performance relative to real-world usage. Distillation from Claude appears to have contributed to some of the performance, and overall the performance is uneven — excelling at certain tasks but not others.

The model is large, slow, and token-hungry, making it difficult for it to replace the role of small, low-cost open models. In direct competition with top closed models, it is expected to lose out in most cases, but it can still be valuable for certain workflows.

Meanwhile, Moonshot AI is reportedly using the buzz around Kimi K3 as a springboard, and has informed investors of plans to pursue a Hong Kong IPO within the next 6 months.