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

Benchmarks from evaluator Vals AI put GPT-6 Sol and Claude Opus 5.5 through coding tasks solo and in teams. The teams burned up to five times as many tokens while gaining almost nothing: just one of four setups improved, and Anthropic's data shows quality leveling off after ten agents. More agents mostly buy speed, says OpenAI's Noam Brown, not better output.

Summary written with AI by HowTechMoves, based on reporting by The Decoder.