Our Take
SandboxAQ is betting the bottleneck in drug discovery AI is access, not accuracy—and it may be right.
Why it matters
Computational chemists at pharma firms have tried other software and hit walls. If Claude + LQMs actually clears those walls without requiring local infrastructure, adoption expands to labs that couldn't justify the engineering overhead. The $950M Alphabet spinout is targeting a $50+ trillion quantitative economy (per company claim)—and picking the interface over raw model performance as its wedge.
Do this week
Computational chemists at large pharma or materials firms: request early access to the Claude integration this week so you can run a real molecular simulation against your current workload before committing to a new vendor contract.
SandboxAQ and Anthropic integrate drug discovery tools into Claude
SandboxAQ, the quantum and AI company spun from Alphabet and chaired by Eric Schmidt, has embedded its large quantitative models (LQMs) directly into Claude. The integration removes the infrastructure requirement that previously forced researchers to run SandboxAQ's models on their own computing systems.
SandboxAQ's LQMs are built on physics equations and real-world lab data, not text patterns. They simulate molecular dynamics, quantum chemistry, and microkinetics—how candidate molecules behave before anyone enters the lab. The company raised over $950 million (company-reported) and counts pharmaceutical researchers, materials scientists, and computational chemists as customers, typically at large industrial firms hunting for marketable products.
The Claude integration exposes these models through natural language. No specialized infrastructure. No PhD in computing required. Users can ask the model about molecular candidates and receive simulations without provisioning servers or rewriting workflows.
Interface, not accuracy, is the real constraint
Competitors like Chai Discovery and Isomorphic Labs have raced to build better models. SandboxAQ's bet is different: the models are already good enough. The customers who matter—researchers at Pfizer, Merck, or materials firms—have tried other software. It didn't work. Not because the science was weak, but because deployment broke in translation to the real world.
A researcher today needs to own or rent computing infrastructure, learn a specialized API, debug integration, and maintain version control. SandboxAQ is removing three of those friction points by moving the entire interaction into Claude, a tool many labs already use for writing, analysis, and reasoning.
The target market is explicit: companies pursuing quantitative problems—biopharma, financial services, energy, advanced materials (per company claim of $50+ trillion addressable market). For those labs, a conversational interface to physics-grounded models could collapse months of engineering into weeks. Whether it actually does depends on how well the Claude integration handles domain-specific reasoning and failure modes real molecules present.
Test the integration against your actual molecules
If you run a computational chemistry or materials science team, the Claude + SandboxAQ integration is worth a trial, but only if you bring a real problem. Academic benchmarks will not tell you whether this saves you two weeks or two hours. Run your next candidate molecule through Claude against the old workflow and time the difference. Count integration bugs, hallucinations, and cases where the model gives you plausible but wrong physics. Document which classes of molecules the conversational interface handles well and which ones require you to drop back to the command line or custom code. That test will tell you whether you can retire your existing tools or whether SandboxAQ remains a specialist second opinion.