The spectral
foundation model.
Early access, open now.

Large language models (LLMs) made it possible for anyone to reason with text; the Spectral Foundation Model makes it possible to reason about matter.

Give the SFM a spectral scan…anything from a Raman probe to a near-IR sensor to an FTIR, and it tells you what's there. That means composition, concentration, state, identity without specialized lab infrastructure. Without a priori knowledge of what you’re looking for.

This is the first open access spectral foundation model. We’re releasing it because we want to understand what scientists, researchers and engineers do with it.

Anyone who works with matter. Today that means researchers and lab scientists with spectral data and a question about what's in it. But spectra are everywhere: in drug discovery pipelines, water systems, food supply chains, manufacturing lines, and field sensors. As the SFM absorbs new modalities, it becomes the interpretive layer for all of them, so a process engineer, a farmer, or a first responder can ask the same question a spectroscopist would: what is this? If your work depends on knowing what something is made of, the SFM is being built for you.

Toward one model that reads any spectrum. Every release adds a modality, and every submission makes the model sharper, because scientists confirm or correct its answers and it learns from them in the open. The endpoint is molecular understanding without specialized lab infrastructure or a priori knowledge of what you're looking for: point a sensor at the world, get an answer. Release 001 is the first step, and people using it now are shaping what the model becomes.