New AI Trick Lets One Model Think Like a Team — 20x Cheaper

Big AI chatbots are powerful, but they usually think in one way. To handle tricky problems, companies sometimes use a "multi-agent system" — a team of AI bots, each with a different specialty, that talk to each other until they agree. This works well, but it's expensive: every bot uses computing power, and all that back-and-forth conversation adds up fast.

A new research paper proposes a smarter approach called Mixture of Roles, or MoRe. Instead of running many separate bots, they train a single model to hold multiple "roles" — essentially many minds inside one brain. When you ask it a question, it quickly blends the right mix of those minds and answers in a single step. It's like hiring one consultant who has the knowledge of five different experts.

The results are striking. MoRe beat regular single-model AI by 2.2% on reasoning and personality tests. And it matched the accuracy of a full multi-agent team while using 20 times less computing power (measured in "tokens" — the bits of text an AI processes). In plain terms, you get team-level thinking for the cost of a single model.

This is research, not a product yet. But if it scales, the impact is big: companies could offer smarter AI assistants with lower server bills, faster response times, and much less energy use. For everyday users, that could mean AI that feels sharper without costing the earth.