Drug discovery remains one of the most resource-intensive endeavors in modern science, with developing a single viable molecule often requiring over a decade and billions of dollars in investment. Despite advances in artificial intelligence, many startups have focused on refining the underlying models, leaving usability challenges unaddressed for researchers who lack specialized technical infrastructure. SandboxAQ, an Alphabet spinout founded five years ago, is tackling this gap by embedding its proprietary quantitative models into Anthropic’s Claude, enabling conversational access to advanced tools for drug discovery and materials science without requiring users to manage complex computing systems.

The company’s large quantitative models (LQMs) are built on physics-based principles rather than text patterns, allowing simulations of quantum chemistry, molecular dynamics, and microkinetics. These capabilities let researchers predict how candidate molecules might behave in real-world labs before costly experiments begin. With backing from investors including Google’s former CEO Eric Schmidt and over $950 million in funding, SandboxAQ has diversified into areas like cybersecurity while prioritizing accessibility. “For the first time, we have a frontier quantitative model on a frontier LLM that someone can access in natural language,” said Nadia Harhen, the company’s AI simulation general manager. Previously, using SandboxAQ’s models required customers to provide their own infrastructure.

Unlike competitors such as Chai Discovery and Isomorphic Labs, which focus on advancing scientific models, SandboxAQ emphasizes user-friendly tools for computational scientists, research teams, and experimentalists at pharmaceutical and industrial firms. The company’s clients typically turn to SandboxAQ after struggling with existing software that fails to bridge the gap between theoretical predictions and real-world outcomes. By integrating with Claude, SandboxAQ aims to democratize access to AI-driven insights, making complex simulations more attainable for researchers without deep computing expertise.