Kiln model family

Kindling

Status: In development

Small, efficient models for low-resource, low-latency, and specialised local deployment.

Kindling is Kiln’s small and efficient model family. A first checkpoint has been trained and recorded, but it is a model record for development and experimentation, not a public release.

Kindling

Kiln model family

Family direction

Useful where local matters most

  • 01

    Low resource

    Designed for smaller hardware and practical local experiments.

  • 02

    Low latency

    A family direction focused on responsive inference.

  • 03

    Broad foundations

    Kindling remains broadly capable rather than being a single-purpose model.

  • 04

    Recorded lineage

    Every checkpoint keeps its architecture, data, and training record visible.

A small model family with a careful paper trail

The current Kindling checkpoint is not a public release or a user-facing deployment.