AnalysisAI AgentsNovember 3, 2026

Benchmark scores 18 AI models on teamwork; best gets 82.2/100

Read original source →arxiv.org

The Collaborative Intelligence Benchmark tests 240 tasks across 8 teamwork dimensions, drawn from aviation accident reports, clinical cases, coding and negotiations. Frontier models refuse unsafe shortcuts but reach the best deal in only 14% of conflicts, and pass rates fall from 97% to 81% when taking over a task a weaker model cheated on.

How this story unfolded

1 day · 1 report · 2 community posts · 3 of 4 shown

  1. Nov 4
  2. Nov 5

More stories today

Open the live feed