Αἰθήρ · The upper air
Aethel
A foundation model built to reason like a scientist.
Mixture-of-Experts architecture with native multimodal fluency across molecular structures, biological sequences, crystallographic data, and scientific imaging. Built to operate in an iterative research loop, not answer single-shot questions.
The Problem
General-purpose models are breadth-optimized
The more disciplines a model tries to serve equally, the less native fluency it has in any one of them.
The Approach
Scientific reasoning rewards depth
Native fluency in non-text modalities — molecular structures, crystallographic data, biological sequences, time-series signals. These are not text problems with a visual wrapper.
The Result
Depth without sacrificing breadth
Competitive with frontier general models on standard benchmarks. Scientific specialization that doesn't come at the cost of general capability.
~1T
Parameter Class
6
Modalities
5
Scientific Domains
01
Mixture-of-Experts Architecture
Large total parameter capacity — ~1 trillion parameter class — with only ~22B active per query. Depth of knowledge without inference latency cost. Specialized expert modules route dynamically based on query domain.
02
Native Multimodal Scientific Input
Not a text model with plugins bolted on. One architecture trained across modalities from the start — molecular structures, crystallographic data, biological sequences, scientific imaging, time-series signals.
03
Cross-Disciplinary Reasoning Core
Chemistry, materials science, biology, earth sciences, and mathematics addressed through one shared reasoning foundation. A single query can pull organic chemistry + materials prediction + biological sequence understanding together.
04
Iterative Agentic Operation
Designed to work inside a loop — propose hypothesis, execute or observe, score outcome, keep or revert — not just answer single-shot questions. This is what separates a model that answers questions about science from a system that does science.
05
Research-Grade Math and Logic
Complex multi-step problems at the level of graduate qualifying exams and olympiad-style reasoning. Symbolic manipulation, proof construction, constraint satisfaction.
06
Strong General Reasoning Retained
Scientific specialization doesn't come at the cost of being broadly capable. Competitive with frontier general models on standard reasoning and knowledge benchmarks.
Aethel is part of the Morbius research infrastructure — the foundation model layer that enables research agents to operate with scientific fluency.