MNIST Curves Suggest a Single Agent Needs 10-100x Budget to Match a Team of Agents

DimitrisPapail · x · 2026-09-22

DimitrisPapail adds to the multi-agent vs single-agent scaling discussion: from the MNIST curves, one can infer an upper bound — it takes 10-100x more token budget for a single best agent to reach the team's performance level.

This provides further quantitative support for team-of-N outperforming single-agent-with-more-compute, consistent with ARC-AGI-3 ablation results.

Related event: Multi-agent vs single-agent scaling debate: teams win with 10-100x efficiency edge(6 posts)→

Original post →

More from coding & agent

coding & agent channel →