Kimi K3 arrives on Modal with shared API access

Kimi K3 is Moonshot AI’s open-weights flagship model on Modal, built as a 2.8-trillion-parameter Mixture-of-Experts system with native vision and a 1M-token context window. Moonshot describes it as an open 3T-class model aimed at long-horizon coding, knowledge work and reasoning.

The model activates 16 of its 896 experts for each token and uses Moonshot’s Kimi Delta Attention architecture. The released weights use MXFP4, with MXFP8 activations.

Modal offers Kimi K3 through a live OpenAI-compatible Shared API, allowing existing SDKs to point to the endpoint. Shared API pricing is billed per million tokens: prompt tokens cost $3.00 /MTok, cached prompt tokens cost $0.30 /MTok, completion costs $15.00 /MTok and reasoning costs $15.00 /MTok.

Developers can also deploy Kimi K3 as an Auto Endpoint, with Modal tuning and autoscaling the inference stack. Auto Endpoint compute is billed by the second and only while it runs, with support for bringing fine-tuned weights.