Chinese artificial intelligence startup Moonshot AI unveiled its latest large language model, Kimi K3, on Thursday, entering a fiercely contested global race among frontier AI developers. The model was trained on 2.8 trillion parameters, signaling a major escalation in computational scale.
Kimi K3 matches the performance of Anthropic's Claude Fable 5 but trails OpenAI's GPT-5.5 and GPT-5.6 Sol models on overall benchmarks, according to the company. However, the model ranks first globally in website programming tasks, per Arena AI evaluations.
The announcement sent shockwaves through China's AI sector, with shares of domestic rivals Z.ai and Minimax plunging as much as 28 percent. The market reaction underscores the stakes for Moonshot, which raised $2 billion at a valuation exceeding $20 billion in May.