Accurate.
Affordable.
A compact model that runs cheaply and answers correctly is the goal, not size for its own sake.
Glint Research. Intelligence, Unbound.
Independent research.
Compact models with a purpose.
We are changingdirection.
The small language model race is behind us.
Glint Research is moving toward compact, inspectable models for biology, robotics, cybersecurity and medicine.
From a race
to a purpose.
A wider scope.
Deeper questions.
-
01
Finish the
current line.A last group of releases is already in motion. Once it is out, the old model series is discontinued. Its records stay public.
-
02
Make room
for research.Work in a wider scope, with deeper projects and outside collaboration. No release dates yet.
-
03
Models that
help people.Compact models for fields where better information can change an outcome.
Research in the real world.
Four fields of inquiry.
Safety across all of them.
01 / 04
Biology
Compact models for understanding biological complexity. A direction for deeper research and outside collaboration.
Research directions.
No release dates yet.
Small where
the task allows.
What we expect
from a model.
Accurate.
Open to
inspection.
We want to see how a model works and why it fails, not just that it scores well.
Failures,
published.
Negative and broken results go on the record next to the wins. A model has to be understood before it is trusted.
Earlier work.
Still open.
Models from the small language model race remain on Hugging Face as a public record. We’re finishing the current line. The work stays accessible.
Research, evaluation, data, feedback.
There is room for yours.
What do you
want to
build?
We are a small team.
Most of the conversation happens on
Discord.