Moonshot describes Kimi K3 as a 2.8 trillion-parameter model with native visual capabilities and a one-million-token context window. Its announcement also says that the model uses Kimi Delta Attention and Attention Residuals. Those are the company’s descriptions; they should be read alongside the technical report when it is published.
For developers, the useful first questions are narrower than a launch claim: which model identifier is available, what the API documentation lists for pricing and modes, and when the published weights and additional evaluation details are due. The official announcement and quickstart are the sources for those details.
This page is intentionally limited to information Moonshot has made public. It does not turn provider benchmarks into independent rankings, and it does not infer deployment requirements that have not been documented.