Quantum Data Framework for Physical AI.

Sirius Quantum is the quantum data framework for Physical AI — an on-premise quantum data relabelling engine, an open simulation network, and open-source quantum datasets and benchmarks spanning quantum chemistry, molecular biology, drug development, materials discovery, and custom-built quantum finance.

• Zilver: a distributed quantum simulator for integrated GPUs. Apache 2.0.

Canis M is the first molecular model trained on quantum tokens: compact representations of a molecule's electronic structure, read by the model in place of atomic coordinates. Trained on 50 molecules, it is more accurate than the standard approach trained on 16,000.

• Model: SiriusQuantum/canis-m
• Paper: Tokenising quantum data for label-efficient training of molecular models
• Announcement: siriusquantum.com/canis-m

We are pioneering quantum-native data for AI. We take the benchmarks the field already trains on, including QM9, QM7b and MoleculeNet, and relabel them with exact quantum computation instead of classical approximations. We also release new datasets built from quantum data from the start, including the first public labelled dataset for barren plateau research. All are open on Hugging Face, across quantum chemistry, drug discovery, quantum physics and finance.