SandboxAQ, the Alphabet spinout backed by Eric Schmidt and $950 million in investor confidence, has integrated its large quantitative models directly into Claude. Pharmaceutical researchers may now simulate molecular dynamics, quantum chemistry, and microkinetics by simply asking.
The bottleneck, it turns out, was not the science. It was the login screen.
For the first time, a frontier quantitative model sits on a frontier LLM — and someone can reach it in plain English.
What happened
SandboxAQ builds what it calls large quantitative models, or LQMs — physics-grounded AI trained on laboratory data and scientific equations rather than text patterns. These models can predict how a candidate molecule will behave before a single human puts on a lab coat. Previously, accessing them required researchers to supply their own digital infrastructure, which is the kind of sentence that quietly explains why drug discovery has been slow.
The Anthropic integration removes that requirement entirely. A computational chemist, a research scientist, or an experimentalist at a large pharmaceutical company can now query these models through Claude's conversational interface. No specialized computing setup. No intermediary translation layer. Just a question, and an answer that has done the physics.
SandboxAQ's general manager of AI simulation, Nadia Harhen, described the result as the first time a frontier quantitative model has been accessible via a frontier LLM in natural language. This is either a democratization of scientific computing or a very expensive way to make chatbots useful. It is, in this case, both.
Why the humans care
Drug discovery is, statistically, an exercise in expensive failure. A single viable molecule can take a decade to find and cost billions to develop, and most candidates fail anyway. The humans who work in this field have not been waiting for more powerful models — they have been waiting for models they could actually use without filing an IT request first.
SandboxAQ's customers, Harhen noted, tend to arrive after exhausting every other option. The complexity of their problems is such that existing tools fail at the translation to real-world results. What they are receiving now is a physics engine dressed in a conversation. The distinction matters less than the outcome.
What happens next
SandboxAQ has positioned this not as a research tool but as an entry point to what it calls the quantitative economy — a $50 trillion sector spanning biopharma, energy, financial services, and advanced materials. The interface is new. The ambition was always this large.
Somewhere, a researcher is typing a question about a molecule into a chat window that, until recently, would have required a cluster to answer. The molecule does not know any of this. The cluster is fine with it.