Kimi K3 tops global rankings with 2.8 trillion parameter model

Moonshot AI has released Kimi K3, a large language model with 2.8 trillion parameters, which has achieved the top position in several industry benchmarks including LiveCodeBench, AIME, and MATH-500. Kimi K3 outperforms GPT-4o and other leading models in these evaluations. The model's architecture employs Mixture of Experts (MoE) with high sparsity and advanced reinforcement learning with Long Chain-of-Thought (Long-CoT) to enhance reasoning and coding abilities. In the new global ranking of large language models, Kimi K3 was the highest-rated Chinese model. Moonshot AI emphasizes that this is just the beginning of their innovation journey.