SandboxAQ Brings Physics-Grounded Drug Discovery Models to Claude

SandboxAQ has integrated its scientific AI models directly into Claude, giving researchers access to quantum chemistry calculations and molecular dynamics simulations through a conversational interface that requires no specialized computing setup. The move puts a class of physics-grounded tools, previously limited to teams with substantial in-house compute, inside the same chat window that millions of Claude users already open daily.

What Large Quantitative Models Actually Do

SandboxAQ calls its models large quantitative models, or LQMs, a label meant to mark them out from the large language models that power most conversational AI. The distinction is technical and material. LQMs are physics-grounded: rather than learning from patterns in human-written text, they are built on the equations that govern molecular behavior. They can run quantum chemistry calculations, simulate the dynamics of molecules as they move and interact over time, and model microkinetics, the study of how chemical reactions proceed at the level of individual molecules.

SandboxAQ, founded roughly five years ago as an Alphabet spinout, has Eric Schmidt, Google's former chief executive, as chairman. The company has raised more than $950 million from investors. Its LQMs have been in use within pharmaceutical research for some time. What the Claude integration changes is who can reach them and at what cost.

Removing the Infrastructure Bottleneck

Researchers in drug discovery often point out that the binding constraint on AI-assisted work is not model quality. The models already exist. The constraint is logistics: building and maintaining the infrastructure to run them, handling API authentication, managing versioning, and ensuring that a biologist who is not a software engineer can still get results. That friction slows adoption and concentrates access within a narrow tier of well-resourced institutions.

Previously, users of SandboxAQ's LQMs had to provide their own digital infrastructure to run the models. The Claude integration removes that requirement entirely. A researcher can now open Claude's interface, describe a molecular simulation they need, and receive results without standing up a single server. Claude's conversational layer also handles result interpretation, reducing the gap between a raw simulation output and a biological insight that a non-specialist can act on.

"The move sidesteps the biggest bottleneck in AI drug discovery — not building better models, but getting them into the hands of researchers who don't have PhDs in computer science." TechCrunch, May 18, 2026

What's Coming Next

SandboxAQ and Anthropic have confirmed that additional drug discovery models will become available through Claude in the coming months. AQPotency is designed to identify and rank drug candidates computationally, helping research teams concentrate synthesis and testing efforts on the molecules most likely to produce useful compounds. AQCell goes a step further, simulating cellular responses to candidate compounds and predicting potential toxicity risks before any laboratory testing occurs.

Together, those additions would place a meaningful portion of a standard computational drug discovery pipeline within Claude's interface. A medicinal chemist, pharmacologist, or research scientist at a smaller institution, one that cannot afford a team of computational biologists and the infrastructure they require, could in principle run much of the same analysis that has historically required dedicated teams and specialized hardware.

The pattern fits Anthropic's broader enterprise approach. Rather than asking partners to build integrations from scratch, Anthropic has been pulling domain-specific capabilities into Claude's interface directly. The Bristol Myers Squibb partnership moves Claude into pharmaceutical research workflows at the enterprise level; the SandboxAQ deal adds depth to what Claude can do inside those workflows by embedding physics-based simulation capability alongside language understanding.

The Broader Implication for Scientific Research

The question for drug discovery AI has never been whether powerful simulation tools could exist. They already did. The question was always who could reach them. Quantum chemistry simulations and molecular dynamics engines have been available to academics and large pharma companies for years. Making them accessible through a conversational interface that requires no specialized infrastructure is a different kind of advance: it expands the population of researchers who can participate in computational drug discovery, not just the capabilities available to those who already could.

It also positions Claude as more than a writing or coding assistant within scientific contexts. Embedding LQMs alongside Claude's language capabilities means a researcher can move, in a single session, from reading a paper on a target protein to running a simulation of how a candidate molecule interacts with it to drafting a summary of the results. That workflow compression is, arguably, more consequential for scientific productivity than any individual improvement to a single model's accuracy. Learn more about Anthropic's research agenda and the principles behind how Claude's development is guided.