Research: AI agent teams cost up to 5.1x more for tiny quality gains

Vals AI research finds AI agent teams barely outperform solo agents while costing up to 5.1x more; only one of four tests with GPT-6 Sol and Claude Opus 5.5 showed measurable gains. Anthropic data also shows quality plateauing beyond ten agents as token costs keep rising.

• OpenAI and Anthropic Cut AI Model Prices, Escalating Price-Performance Race2026-09-23
• Claude Opus 5.5 and GPT-6 Launch Spark Price War2026-09-23
• Developer uses Claude to clone Adobe's software suite2026-10-07
• Anthropic Releases Claude Haiku 5.52026-10-07
• A Series of Unfortunate Events for OpenAI Users2026-10-06