Evals company Vals AI has found that AI agent teams deliver almost no better results than single agents while wasting massive amounts of tokens. In recent tests evaluating GPT-6 Sol and Claude Opus 5.5 on the Vibe Code Bench, team setups cost between 1.8x and 5.1x more than single agents. Out of four comparisons between teams and solo agents, only one showed a statistically significant improvement, which was GPT-6 Sol scoring 7.3 points higher at medium reasoning. At maximum reasoning effort, the team configuration provided neither model with any real advantage.
Additional tests reinforce these findings regarding scaling limits and cost inefficiencies. Anthropic observed that quality gains shrank when adding more agents with Opus 5.5, and while larger teams reached performance levels faster, going from ten to 100 agents only nudged scores up slightly. On a knowledge base task, Fable 5.1 scores actually dipped slightly when scaling from 30 to 100 agents. OpenAI researcher Noam Brown confirmed that multi-agent systems primarily buy speed rather than better quality, noting that tasks like web research parallelize well while creative tasks do not. OpenAI developer Eric Provencher also warned against agent swarms, describing them as wasted money due to coordination breakdowns.
This research matters because it highlights severe economic and efficiency limits in current multi-agent artificial intelligence strategies, showing that heavy token expenditure often yields negligible quality improvements.