Moonshot has released Kimi K3, a 2.8 trillion parameter AI model that outperforms Claude Fable 5 on several benchmarks, with open weights and competitive pricing, and China has developed this model using H200s and Huawei Ascend hardware despite US export controls on H800s.
Kimi K3 ranks 2nd in agentic knowledge work, achieving an Elo of 1543 and passing 51% of the rubric, but each task costs $10.57 to run, roughly 10x more than K2.6. The model's high cost is due to its long runs and heavy output, generating more output tokens at a high per-token price.
Kimi K3, an AI model, was given a 3D model and a prompt, and it built a VR companion in approximately 30 minutes, demonstrating its ability to find and deploy local ASR and TTS connected LLM and build VR scenes
Moonshot is a model that may impact spending on Western closed source models due to its expensive and verbose yet effective nature