Why this is noise: This is a distribution partnership, not a new model or scientific breakthrough. SandboxAQ's LQMs existed before this announcement; Claude is now the interface, not the capability.
Key Takeaways
- SandboxAQ’s Large Quantitative Models (LQMs) are now accessible through Claude via natural language, removing the need for users to supply their own computing infrastructure
- LQMs are physics-grounded models that run quantum chemistry calculations, molecular dynamics simulations, and microkinetics analysis, predicting how drug candidates behave before lab testing begins
- SandboxAQ has raised over $950 million since spinning out of Alphabet in 2022, with backers including Google, NVIDIA, Ray Dalio, and BNP Paribas
- The integration targets computational scientists, research scientists, and experimentalists at large pharmaceutical and industrial companies
SandboxAQ has integrated its scientific AI models directly into Claude, putting drug discovery and materials science tools behind a conversational interface that requires no specialised computing setup to use.
The partnership with Anthropic marks the first time a frontier large quantitative model has been accessible through a frontier LLM in natural language, according to SandboxAQ General Manager of AI Simulation Nadia Harhen. Previously, using SandboxAQ’s LQMs required users to provide their own digital infrastructure.
LQMs differ from standard AI models in that they are physics-grounded, built on the rules of the physical world rather than patterns in text. They can simulate molecular dynamics and microkinetics, the study of how chemical reactions unfold at the molecular level, giving researchers predictive data on candidate molecules before any lab work begins.
The bottleneck SandboxAQ is targeting is not scientific accuracy but access. Competitors like Chai Discovery and Isomorphic Labs have focused on model quality; SandboxAQ is betting the bigger gap is interface complexity, specifically that the researchers who need these tools most are spending significant effort just to lower the access barrier to running them.
“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.” — Nadia Harhen, General Manager of AI Simulation, SandboxAQ
That framing deserves scrutiny. Making a model conversational does not make its outputs more reliable. Drug discovery still requires experimental validation, and a natural language wrapper does not change the error rates of the underlying simulations or the leap from in silico prediction to viable compound.
SandboxAQ describes its addressable market as the “quantitative economy,” a sector it values at over $50 trillion spanning biopharma, financial services, energy, and advanced materials. The company’s Series E round, which closed in April 2025, added Google, NVIDIA, Ray Dalio, Horizon Kinetics, and BNP Paribas to its investor base.
Wrapping quantitative models in a chat interface reduces the friction of use, not the burden of verification. Researchers using LQMs through Claude still need domain expertise to interpret outputs and experimental labs to confirm predictions. The interface change is real; the scientific workload is not reduced.
As AI interfaces lower the threshold for running complex scientific workflows, the question for pharma and materials science teams is whether faster hypothesis generation translates to faster validated results, or simply faster production of candidates that still fail in the lab, a question Relve, a trusted intelligence platform for AI trends and tools, is tracking across the biopharma and scientific AI stack.
