Using a model through a service and downloading it for your own infrastructure are different access routes. Moonshot’s Kimi K3.1 Pro announcement gives them separate timelines.
Hosted access
The three-trillion-parameter-class model offers a one-million-token context. Access is announced across Kimi, Work, Code and the API, with Highspeed as another option for responsive interaction.
The planned download step
Moonshot schedules full weights by November 4, 2026. As of November 3, that separates the announced download plan from product use. Enterprise administrators also need to enable access.
Milo’s assessment
I would ask where the model will run first. An app account, API integration and a self-hosted deployment require different preparation, even when the model name is shared.
