SandboxAQ

AI For Science · Materials Chemistry · Enterprise AI

Launch

SandboxAQ makes AQCat catalyst-screening AI available on Claude

It approaches gold-standard DFT accuracy while running orders of magnitude faster, letting researchers skip specialized compute infrastructure entirely.

• SandboxAQ made its AQCat catalyst-screening model generally available on Claude Science via Anthropic's Model Context Protocol on August 19, 2026, moving beyond a May waitlist-based integration.
• AQCat is SandboxAQ's Large Quantitative Model for catalysis; it calculates adsorption energy, the measure of how strongly a molecule binds to a catalyst surface that determines whether a material will work.
• The model is 'spin-aware,' explicitly modeling magnetic behavior so it can accurately screen abundant metals like iron, cobalt and nickel that many machine-learning catalyst models ignore.
• AQCat was trained on 13.5 million high-fidelity DFT calculations across 47,000 catalyst systems and runs up to 20,000 times faster than DFT, versus lab methods that test under 100 candidates a week.
• Beyond Claude, AQCat is also now offered directly through SandboxAQ and on AWS Marketplace, broadening distribution past the earlier limited-access rollout.
• By letting any researcher query catalyst physics in plain English, SandboxAQ is trying to turn a specialist, compute-heavy workflow into a mainstream tool that could reshape R&D economics in green hydrogen, sustainable fuel, fertilizer and plastics recycling.