China’s AI scene has a habit of sending Western tech investors scrambling for their sell buttons. Moonshot AI, a Beijing-based startup founded by Yang Zhilin, just did it again.
The company unveiled Kimi K3 on July 16-17, 2026, billing it as the world’s largest open-weight AI model at launch, built on a jaw-dropping 2.8 trillion parameters. For context, more parameters generally means a model has more capacity to learn complex patterns, though raw size is not the whole story.
What Kimi K3 actually does
Independent evaluations placed Kimi K3 above OpenAI’s GPT-5.6 on certain tasks, while Anthropic’s Claude Fable 5 edged it out overall. The model showed particular strength in front-end coding, which is exactly the kind of practical capability that makes enterprise buyers pay attention.
Moonshot also revealed the pricing structure for its API: $3 per million input tokens and $15 per million output tokens. The model supports a context window of 1 million tokens, meaning it can process and reason over very long documents or conversations in a single pass. In English: you can feed it an entire book and it won’t forget the first chapter by the time it gets to the last.
