Two divisions.
One architecture.

AENEA Quartz is the primary division of AENEA Global — a vertically-integrated AI infrastructure operation that unifies our data substrate (Quartz), our tokenizer research (QT), and our model laboratory (AENEA) under one engineering roof. Everything else the group builds is downstream of this layer.

The Quartz data pipelines ingest, sanitise, and validate the raw corpora used to train our models. The flagship QT-Cittern-12k tokenizer is calibrated for edge routing — a 12,000-entry vocabulary that strikes the optimal balance between compression efficiency and routing head compactness, keeping the embedding projection aligned to Tensor-Core tile sizes while preserving strong coverage across code, bash, and human languages.

The AENEA model laboratory designs small, dense, highly-efficient architectures built specifically for factual recall and deterministic classification. The Pinta family has demonstrated that when training geometry is aligned with ultra-clean data, parameter counts can be reduced by orders of magnitude without sacrificing routing accuracy. The 24-layer Cittern-2.0 architecture — block Householder orthogonal attention, parallel residuals, RoPE positional embeddings — sits on top of the Cittern-2.0 base checkpoint, fine-tuned with orthogonal regularisation to preserve the Stiefel manifold geometry.

Open source · Apache 2.0 · 12K vocab · 92.45% routing accuracy