Overview
- SandboxAQ announced general availability of AQPotency and AQCat on Wednesday, making both models accessible through Claude via the Model Context Protocol and on SandboxAQ’s site with AQCat added to AWS Marketplace and Google Cloud listing planned.
- AQPotency is a structure-agnostic Large Quantitative Model that ranks molecule–target pairs in seconds on standard hardware, returns confidence estimates for each prediction, and has been used in eight customer programs with company-reported experimental follow-up.
- AQCat is a spin-aware materials model trained on the AQCat25 dataset of 13.5 million high-fidelity DFT calculations and is claimed to recover near-DFT adsorption energies for magnetically complex catalysts while running thousands to up to 20,000 times faster than traditional DFT.
- Both tools are delivered through a plain-language interface meant to lower technical barriers so smaller labs and students can pre-screen large design spaces before committing to costly DFT runs or lab assays, which could speed early-stage discovery and cut experimental costs.
- The performance claims are presented by SandboxAQ and supported by academic users quoted in the announcements, but independent peer-reviewed benchmarking and broader validation are still needed to confirm real-world limits and reliability.