Intelligence should be plural

There is no universally best model.

That sentence is the reason this lab exists. Intelligence is dependent on the task, the environment, the person, and the objective. As models become more capable, understanding those differences will become as important as measuring raw capability.

Most of the field still treats intelligence as a single axis. Leaderboards collapse judgment, style, caution, creativity, and usefulness into one number. That is a useful simplification for some questions. It is a poor description of how people actually experience AI.

We are building Last Labs around a different bet: that the future has thousands of models, not a handful. Some will reason differently. Some will write differently. Some will specialize. Some will be more creative, empathetic, precise, opinionated, cautious, or strange.

Those differences should be measurable, understandable, and available to people.

That means benchmarks that capture what current evaluations miss. Environments where models can be evaluated and trained through interaction. And open infrastructure for routing, tracing, experimentation, and learning across models.

AI should not reduce humanity to a single interface with a single kind of intelligence. It should expand what people are capable of.