AivexaNewsSearch
AI news for builders and product teamsChecked every hour

AI agent teams waste massive tokens for barely measurable quality gains, research finds

Collected Oct 11, 2026

Vals AI has published research finding that teams of AI agents deliver only slight quality improvements over a single agent while using significantly more tokens, according to The Decoder. The work examined multi-agent setups and reported that coordination costs scale far faster than results.

The headline figure: agent teams cost up to 5.1x more tokens, yet the quality gain is barely measurable. Only one out of four tests with GPT-6 Sol and Claude Opus 5.5 produced a measurable improvement. The article also cites Anthropic's own data, which reportedly shows that beyond ten agents, quality plateaus while token costs continue to climb.

This is a notable result for teams building agentic workflows. Orchestrating multiple agents adds overhead — routing, handoffs, retries and duplicated context — and the data suggests those costs are not repaid in better output past a fairly low agent count. A likely trade-off for developers is that scaling agent count is not a reliable path to quality; capping team size and investing in a stronger single agent may be more cost-effective. These points are my own reading of the reported figures, not claims from the research itself.

Why it matters: teams designing multi-agent pipelines should treat agent count as a cost knob rather than a quality lever. If the findings hold, products relying on large agent teams may be paying substantially more for little measurable benefit, and evaluation should focus on where extra agents actually change results.

Read at The Decoder

Based on reporting from the original publisher. Visit the source for full context and later updates.

Publisher excerpt

Teams of AI agents barely outperform solo agents but cost up to 5.1x more, according to Vals AI. Only one out of four tests with GPT-6 Sol and Claude Opus 5.5 showed a measurable gain. Anthropic's own data backs this up: beyond ten agents, quality plateaus while token costs keep climbing. The article AI agent teams waste massive tokens for barely measurable quality gains, research finds appeared first on The Decoder .