OSC700: A GW+BSE Dataset of 700 Organic Semiconductor (OSC) Crystals with an LLM-Assisted Data Extraction Agent

Key parameters for organic optoelectronic device applications derive from the excited-state properties of molecular crystals. Data scarcity (both experimental and theoretical) limits the ability to train machine learning (ML) models to predict these properties. We present the OSC700 dataset of excited-state properties of 700 crystalline organic semiconductors (OSCs) with up to 1168 atoms in the unit cell, calculated using state-of-the-art many-body perturbation theory within the GW approximation and Bethe-Salpeter equation (GW+BSE). OSC700 contains crystals of compounds comprising C,H,N,O,F,S capable of forming a variety of molecular scaffolds and crystal packing motifs. OSC700 is larger and significantly more diverse than the earlier PAH101 dataset [Scientific Data 12, 679 (2025)]. The data records contain the fundamental gap, quasiparticle band structure, static dielectric constant, first singlet exciton energy (optical gap), first triplet exciton energy, full dielectric function, and optical absorption spectra for light polarized along the three lattice vectors. Because absorption experiments on molecular crystals are rare (as opposed to solution or amorphous phases), we developed an LLM-assisted data extraction agent in order to find and retrieve available optical gap data for the materials in OSC700. The agent was able to find experimental data for 16 materials, which we used to validate the results of our calculations. The OSC700 dataset can be used to train ML models to predict the excited-state properties of crystalline organic semiconductors and find promising materials for optoelectronic device applications.