Key Takeaways
- SandboxAQ's physics-based AI models now run inside Claude, accessible through natural language
- The integration removes the need for specialized computing infrastructure to run molecular simulations
- SandboxAQ has raised over $950 million and is backed by former Google CEO Eric Schmidt
The Problem Isn't the Model
Drug discovery remains one of the most expensive failures in business. Finding a single viable molecule takes roughly a decade and costs billions of dollars. Most candidates still fail. AI startups have spent years promising to fix this, but their tools require technical sophistication that most researchers don't have.
SandboxAQ thinks the bottleneck isn't the science. It's the interface. The company has partnered with Anthropic to put its drug discovery and materials science AI directly inside Claude. Researchers can now run complex simulations by typing questions in plain English.
What Are Large Quantitative Models?
SandboxAQ builds what it calls large quantitative models, or LQMs. Unlike language models trained on text patterns, LQMs are built on the rules of physics. They run quantum chemistry calculations and simulate molecular dynamics and microkinetics, which is how chemical reactions unfold at the molecular level.
This matters because it tells researchers how candidate molecules will behave before anyone steps into a lab. If a molecule is likely to fail, researchers find out early rather than after years of expensive trials.
“For the first time, we have a frontier [quantitative] model on a frontier LLM that someone can access in natural language.”
— Nadia Harhen, General Manager of AI Simulation at SandboxAQ
The Infrastructure Barrier
Until now, using SandboxAQ's models meant building your own computing infrastructure. That requirement limited the customer base to computational scientists and research teams at large pharmaceutical or industrial companies, people already equipped to handle the technical overhead.
The Claude integration removes that barrier. A chemist can type a question about molecular behavior and get simulation results without writing code or provisioning servers. SandboxAQ is betting that accessibility, not just accuracy, is what separates useful AI from impressive demos.
The company's customers typically arrive after trying other software and hitting walls. As SandboxAQ describes it, these are people dealing with problems complex enough that standard tools failed to translate into real-world results.
Competing on Accessibility, Not Just Science
Well-funded competitors like Chai Discovery and Isomorphic Labs have focused on building better models. SandboxAQ is taking a different approach by focusing on who can actually use them.
SandboxAQ frames its target market as the "quantitative economy," a $50 trillion sector spanning biopharma, financial services, energy, and advanced materials. That language signals ambition beyond drug discovery. The company isn't building a chatbot. It's chasing the sectors AI is supposed to transform.
How AI providers are managing access and usage limits
SandboxAQ's Broader Business
SandboxAQ spun out of Alphabet roughly five years ago. Eric Schmidt, Google's former CEO, serves as chairman. The company has built multiple business lines, including a cybersecurity division alongside its scientific modeling work.
The $950 million in funding gives SandboxAQ runway to pursue multiple verticals simultaneously. Integrating with Claude suggests a platform strategy: rather than building its own consumer-facing products, SandboxAQ is embedding its capabilities where scientists already work.
Logicity's Take
What This Means for Drug Discovery
The integration could shrink the gap between computational scientists and bench researchers. Lab scientists who previously needed to request simulations from a specialized team can now run them directly. That compresses feedback loops and could accelerate early-stage candidate screening.
The bigger question is whether natural language interfaces are precise enough for scientific work. Ambiguity that's fine in casual conversation could produce misleading results in molecular simulation. SandboxAQ will need to show that accessibility doesn't come at the cost of accuracy.
Frequently Asked Questions
What is SandboxAQ?
SandboxAQ is an Alphabet spinout focused on scientific AI and cybersecurity. Eric Schmidt, former Google CEO, serves as chairman. The company has raised over $950 million.
What are large quantitative models (LQMs)?
LQMs are AI models built on physics rather than text patterns. They run quantum chemistry calculations and simulate molecular behavior, helping researchers predict how drug candidates will perform before lab testing.
How does the Claude integration work?
Researchers can access SandboxAQ's simulation tools through Claude's conversational interface. They type questions in natural language and receive molecular simulation results without needing specialized infrastructure.
Who are SandboxAQ's main competitors?
Chai Discovery and Isomorphic Labs are well-funded competitors focused on building better drug discovery models. SandboxAQ differentiates by focusing on accessibility rather than just model performance.
Need Help Implementing This?
Source: TechCrunch / Lucas Ropek
Anthropic Launches Self-Hosted Sandboxes and MCP Tunnels for Claude Agents
Anthropic has introduced self-hosted sandboxes and MCP tunnels for Claude Managed Agents, allowing enterprises to execute agent tools on their own infrastructure and securely connect to internal databases via the Model Context Protocol. These features provide the security and infrastructure framework that supports the deployment of advanced agents like SandboxAQ in corporate environments.
Huma Shazia
Senior AI & Tech Writer
Produced with AI assistance and reviewed by the Logicity editorial team. Learn more in our Editorial Policy.
