Models

What we publish

Requants, hardware-specific builds, and our own fine-tunes of the models we run ourselves, published on HuggingFace. Two model families so far; more will land here as we build and quantize them.

Illustration of a llama peeking out of a small server box, with a glowing halo above its head

Signal 3.8·27B

Qwen3.8-27B that gets to the answer faster. 57% fewer answer tokens and 52% fewer thinking tokens than the base model, at matching or better answer quality.

Qwen3.8-Flash-Next

Seven mainline-compatible tiers plus a Strix Halo build, each benchmarked against the leading community builds at matching sizes. Our smallest tier beats the field in its class.