Multi-agent teams cost more but deliver little quality gain
What happened
On 11 October 2026, evaluation firm Vals AI tested GPT-6 Sol and Claude Opus 5.5 on Vibe Code Bench, running each as a single agent and as a multi-agent team, with reasoning effort set to medium and maximum. The results show that multi-agent teams consume large amounts of tokens, costing 1.8 to 5.1 times as much as a single agent, while quality barely improves. The report notes that this imbalance between cost and benefit is the main problem facing multi-agent collaboration today.
Written by AI from the reports · updated 5 d ago
Timeline
Follow the reports to see the event from each side.
- The DecoderAI agent teams waste massive tokens for barely measurable quality gains, research finds
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.
Heat of this event
Current heat 3·Peak in the comparable range 9（Oct 12 03:00）·Change over 24 hours in the comparable range -51%
The trend compares only the same accounts observed without a break, so its range can be narrower than the current heat count. Move the pointer or tap the chart to see each hour; with a keyboard, use the left and right arrow keys.