Why Do Four AI Agents Finish Some Problems Twice as Fast?

Noam Brown says four agents finish some benchmark problems twice as fast as one, because the four of them work for half as long. The cost is about twice as much for an answer twice as quickly. At 16 agents the same pattern continues, a little less efficiently. He calls the speedup slightly sublinear, and it depends on the problem. Math is very parallelizable. Searching the web for a deep research report is extremely parallelizable. Writing a novel is not: 10,000 agents would probably help no more than 10,000 people trying to write a novel together.

He would not give multi-agent systems even 10 percent of the credit for a Millennium Prize effort. A strong general-purpose model that can work for a long time and think in parallel is the core reason that effort is possible. Multi-agent setups are new and flashy, so they get more of the credit than they deserve. Published plots stop around 16 agents. Runs at the scale of 10,000 agents are too expensive for a careful test. Looking at 64, 128, and 256 agents is the practical next step, and even then it is hard to know what 10,000 agents added over 1,000.