Entity · benchmark

3D-Fit

benchmarkactiveprovisional3d-fit-a2a3c47b·1 events·first seen Jul 21, 2026

Aliases: 3D-Fit

Co-occurring entities

Do Language Models Dream of Binding Molecules? Benchmarking LLMs under Spatial Constraints

More like this (12)

SubFit 3D-RoPE MPI3D SetFit Hyperfitting 4D reconstruction HM3D Hunyuan 3D 3D tracking FLUX3D HumanML3D 3D Medical Image Classification

Recent events (1)

5arXiv · cs.CL·Jul 21, 2026·source ↗

3D-Fit benchmark evaluates LLMs on spatially-constrained drug molecule generation

A new arXiv preprint introduces 3D-Fit, a token-efficient benchmarking strategy for evaluating general-purpose LLMs on structure-based drug design (SBDD) tasks requiring 3D spatial reasoning. The study compares LLMs against specialized diffusion model baselines on pocket-conditioned ligand generation with constraints including anchor fragments, pharmacophore points, and mandatory pocket-ligand interactions. Results show LLMs still lag behind state-of-the-art diffusion models but demonstrate promising ability to handle multiple simultaneous spatial constraints, suggesting a viable path for LLM-based molecular design.

Evaluation and Benchmarking 3D-Fit Do Language Models Dream of Binding Molecules? Benchmarking LLMs under Spatial Constraints