Parallel agent teams improve compute use for better test-time results
Test-Time Compute Scales Better with Parallel Agent Teams
Artificial Intelligence
Summary
When using a fixed computer model to solve hard problems, spending more computing effort during testing can improve results. Common ways either try many independent attempts or one big cooperating group, but both have downsides. The authors show that dividing agents into many small teams and selecting the best team works better. They tested this approach on tasks like programming and proof-solving, where it often beat other methods. A simple math explanation helps understand why mixing teamwork and variety helps.
What this means in practice
- •For software developers: Improve the quality of automated solutions by running multiple small collaborating agent teams during test-time inference to find better outputs.
- •For hardware optimization teams: Optimize long computation tasks like GPU kernel design by using parallel agent teams to get better results without increasing costs.
Authors
Junlin Wang, Haozhe Chen, Ang Li, Rahul Chalamala, Zhuang Liu
Abstract
Test-time scaling improves the performance of a fixed model by spending more computation at inference. There are two common ways to spend it on agents when a verifier can score every attempt. Best-of-N sampling runs many independent agents and keeps the best result but saturates quickly. One large team of collaborating agents can score higher but stops improving as it grows. Can we get the best of both? Our method, parallel agent teams, does so by sampling many small teams in parallel and keeping the best team's result. We test it on 12 long-horizon tasks from GPU kernels and formal proofs to exploit development. With a 27B model and a budget of 100 agents, parallel teams of ten beat parallel agents on 10 of 12 tasks and beat one team of 100 on all four tasks where we ran such large teams. At equal dollar cost the same parallel teams beat a frontier model on half the tasks. A simple statistical model shows that individual agents lose because their many samples have a low mean and that one large team loses because a single sample gets no gain from variance. Sampling many small teams in parallel balances a high mean with the gain from variance. Parallel agent teams thus offer a more effective way to scale test-time compute.