Kairos-10M

Kairos-10M is the smallest public checkpoint in the Kairos family, developed by researchers from Ant Group and ShanghaiTech University. It is positioned as an adaptive zero-shot forecaster, and at roughly 10M parameters it is the lightest-weight way to access the Kairos design while staying entirely in the zero-shot regime.

Architecturally it is an encoder-decoder transformer whose defining feature is adaptivity: the official model cards and project page describe adaptive tokenization paired with instance-adaptive rotary position encoding, so the model varies its patching granularity and positional treatment to match the local information density and temporal structure of each series. It is pretrained on the PreSTS corpus of 300B+ time points, which gives the small checkpoint its broad zero-shot reach across heterogeneous domains.

On TSFM.ai reach for Kairos-10M when you want the adaptive Kairos behavior at the lowest serving footprint and highest throughput, for example when fanning out across many series where per-call cost matters more than peak accuracy. Step up to Kairos-23M for a capacity middle ground, or to Kairos-50M for the highest published capacity in this adaptive family.