---
title: "Foundation models atthe floor of arithmetic."
description: "Fermion Research develops Neutrino local language models, ternary QAT methods, model formats, and inference runtimes for CUDA, Apple silicon, and x86."
canonical: "https://www.fermionresearch.com/"
source: "Fermion Research"
---

![An oil painting of bubble-chamber particle tracks, sage-green spirals and one lime track on a bright cream ground](https://www.fermionresearch.com/images/legacy/bubble-chamber-hero.jpg)

# Foundation models at
the floor of arithmetic.

AI research and models that push intelligence to the physical limits of computation.

## New release

Introducing Phonon-1

Read the announcement

## Mission

Intelligence is becoming the most valuable computation on earth, and the most expensive. Fermion Research exists to change the second part. We build foundation models that push intelligence to the physical limits of computation.

Nature runs a general intelligence on twenty watts. We do not believe that is the floor.

## From the lab

![A field of warm mineral light gathering into a single bright band](https://www.fermionresearch.com/images/launch/phonon-hero.jpg)

AnnouncementAugust 28, 2026

### Introducing Phonon-1

An open speech recognition model for English: a 415 MB download that runs on a laptop or a datacenter GPU and transcribes an hour of audio in about two and a half minutes.

![An amber cellular orbit gathering around a compact central form](https://www.fermionresearch.com/images/backgrounds/amber-orbit.webp)

AnnouncementJuly 27, 2026

### Introducing the Neutrino-1 models

Three open-weight models trained for a compact ternary format and released with CUDA, Apple-silicon, and x86 runtimes.

![A cobalt cellular membrane becoming progressively finer](https://www.fermionresearch.com/images/backgrounds/cobalt-cell.webp)

ResearchJuly 27, 2026

### Intelligence at one-eighth the bits

One-shot two-bit conversion lands near chance. Training inside the constraint produces 72.1 MMLU from a 3.88 GB artifact. This is what changed inside the weights.

![A honey-toned filament carrying energy across a tactile surface](https://www.fermionresearch.com/images/backgrounds/honey-filament.webp)

ResearchJuly 27, 2026

### The Neutrino Engine

An A100 can move the weights for one measured decoder step in 104 microseconds. The step takes 694. The Neutrino Engine is built around the gap.

## Models

## Models

### Neutrino-1 8B

Compact general intelligence for reasoning and tool use.

### Neutrino-1 0.6B

Fast local drafting in a 238 MB model.

## Careers

## Build intelligence at the limits of computation.

Source: [https://www.fermionresearch.com/](https://www.fermionresearch.com/)