Drug discovery ranks among the costliest failures in modern industry. Finding a single viable molecule can take a decade and cost billions, yet most candidates never make it. A wave of AI startups promised to change that—but for the most part, they’ve only eased the burden for researchers already technically adept enough to use their tools. SandboxAQ, however, believes the real bottleneck isn’t the models themselves—it’s the interface.

The company has partnered with Anthropic to embed its scientific AI models directly into Claude, putting powerful drug discovery and materials science tools behind a conversational interface that requires no specialized computing infrastructure. Founded roughly five years ago as a spinout from Alphabet, SandboxAQ counts Eric Schmidt, Google’s former CEO, as its chairman. The company has raised over $950 million from investors and developed multiple business lines, including a cybersecurity division.

What sets SandboxAQ apart, however, is its work on large quantitative models, or LQMs. These proprietary models are “physics-grounded”—built on the rules of the physical world rather than patterns in text. They can run quantum chemistry calculations and simulate both molecular dynamics and microkinetics, the study of how chemical reactions unfold at the molecular level. That matters because it tells researchers how candidate molecules are likely to behave before anyone steps foot in a lab.

“Trained on real-world lab data and scientific equations, LQMs are AI models engineered for the quantitative economy, a $50+ trillion sector spanning biopharma, financial services, energy, and advanced materials,” the company said in a release that makes clear SandboxAQ isn’t building another chatbot or code assistant. It’s chasing the economy that AI is supposed to transform.

While startups like Chai Discovery and Isomorphic Labs—both well-funded bets on better models—have focused on the science itself, SandboxAQ is focused on who can actually use it. Previously, users of SandboxAQ’s LQMs had to provide their own digital infrastructure to run the models. The company’s customers tend to be computational scientists, research scientists, or experimentalists, typically working at large pharmaceutical or industrial companies in search of new materials that can become marketable products.

“Our customers come to us because they’ve tried all the other software out there, and the complexity of their problem is such that it didn’t work or didn’t yield positive results for them when that translation went to take place in the real world,” said Harhen.