Research

How much intelligence survives below floating point is a measurable question. The program is the descent: native training in a ternary-family format, the engine that serves it, and the floor of both.

Community Showcase

Creator runsShowcases

Runs of Neutrino-1 on hardware we do not own: the machine, the decode rate, and the session. Six categories are open.

Explore

Publications

Launch notesResearch posts

The release note and the two posts underneath it: the ternary-family format and the engine that decodes it.

Browse publications

Learn

DocumentationRuntimes

Engine documentation: first-run behavior, checksum gating, the GGUF and MLX paths, benchmark reproduction.

Explore

An amber cellular orbit gathering around a compact central form

AnnouncementJuly 27, 2026

Introducing the Neutrino-1 models

Three models in a proprietary ternary-family format, served by one engine from one artifact on datacenter GPUs, Apple silicon, and desktop CPUs.

Read the announcement

A cobalt cellular membrane becoming progressively finer

ResearchJuly 27, 2026

Intelligence at one-eighth the bits

Native-format training holds frontier capability in a ternary-family format at one eighth the bits of fp16; rounding a trained model to the same depth lands near chance.

A honey-toned filament carrying energy across a tactile surface

ResearchJuly 27, 2026

The Neutrino Engine

The runtime built alongside the format: single-stream decode from one artifact on every backend, with a drafted path verified token for token against plain decode.