Google Cloud adds SandboxAQ's AI models for scientific calculations to its marketplace

Google has connected SandboxAQ's "large quantitative models" (LQMs) to its cloud marketplace. This AI is trained not on internet text but on scientific equations and experimental data. The new models will work alongside Gemini and will be available through the standard Google Cloud Marketplace.

Why LLMs struggle with science

Large language models can do a lot: summarize papers, write code, explain concepts in plain language. But they have a systemic problem with math and precise calculation. They often make errors in computations, confuse physical constants and cannot work with equations the way real researchers need.

The cause is architectural: LLMs are trained to predict the next token in a sequence of text. To them, numbers are symbols, not quantities. A model does not "understand" that 0.001 mol of a substance and 1 mol differ by orders of magnitude. In chemistry, physics or biology, that kind of error can mean a failed experiment.

"Language models sound convincing when they write about science, but

that is not the same as calculating correctly."

What SandboxAQ's LQMs are

SandboxAQ is a company spun out of Alphabet (Google's parent company) in 2022. It focuses on quantum simulation and AI for scientific tasks. Large Quantitative Models are a fundamentally different class of AI: instead of predicting text, the models are trained to work directly with mathematical structures and numerical data.

LQMs are trained on structured scientific datasets:

• Equations from physics, chemistry, materials science and biology
• Laboratory data and experimental results
• Numerical methods and algorithms for modeling
• Parameters of molecular structures and material properties

The result is models that solve computational problems with an accuracy a general-purpose LLM cannot reach.

How the pairing with Gemini will work

Combining LQM and Gemini opens fundamentally new ways for enterprise customers to use cloud AI. Researchers used to have to choose: either precise but highly specialized scientific tools, or general-purpose LLMs that speed up work with text but are unreliable at calculation. Now the roles are clearly split: Gemini handles text context, frames the task and explains results, while the LQM is responsible for exact computation.

Potential applications include developing new drug compounds, climate modeling, designing materials for electronics, and quantum simulation. Access through Google Cloud Marketplace means customers can use LQMs without deploying their own infrastructure.

What this means

Adding specialized "science" AI to Google Cloud acknowledges that no single general-purpose model solves every task. The era of "one LLM for everything" is gradually giving way to a mosaic of specialized tools. For research organizations, this opens the possibility of bringing AI into lab workflows without trading speed for accuracy.

Google, for its part, is extending Cloud beyond enterprise automation toward science and deep tech.