We set up a multi-agent POMDP environment for six frontier models to work through a clinical case with two constraints: 1. A limited budget. 2. A patient who isn't aware of her complete medical condition. The HealthBench leader (Qwen) performed the worst. Thread 🧵
Six Frontier Models Tested in Multi-Agent Clinical POMDP
judgement_labs built a multi-agent POMDP environment where six frontier models worked through a clinical case under limited budget and an uninformed patient. HealthBench leader Qwen performed worst.