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.