Robot-Use LLM Agents
Robot-Use LLM Agent Evaluation
We test how well today’s frontier LLMs work as robot-use agents.
An LLM can use a robot the way it uses a calculator, a search engine, or a computer: as a tool. We put existing frontier models in control of real robots and measure what they can and cannot do.
The tests run on physical robots, across a range of everyday tasks, so the results reflect what a model does in the real world, not only in simulation.