AI代理團隊消耗大量token卻只帶來微小質量提升
研究發現AI代理團隊成本是單個代理的1.8倍至5.1倍，但質量提升卻非常有限。這將影響使用AI執行任務的公司。
vals ai tested gpt-6 sol and claude opus 5.5 on the vibe code bench in solo and team setups at two reasoning levels. teams cost 1.8x to 5.1x more than single agents but only showed a statistically significant improvement in one case: gpt-6 sol at medium reasoning with a 7.3 point higher score. anthropic's own tests with opus 5.5 showed similar results. openai researcher noam brown noted that multi-agent systems mainly buy speed, not better quality. this affects companies using ai for tasks and may signal a need to reevaluate the use of agent teams.