Ping-pong decision test: Jev scores 24/24, beating larger models

this tiny ai model is making expensive ones look absolutely ridiculous four models playing ping-pong, making split-second decisions. and somehow the models built for simple decisions are destroying the ones built to think. [look at these results] → TypeSafe Jev: 24/24 hits → Microsoft Decision-1: 24/24 hits → Qwen3.8 27B: 13/24 hits → Haiku 5.5: 6/24 hits it's a simulated test, but the cost comparison gets even more interesting. → Jev: $12.60 per million decisions → Qwen: $82.50 → Haiku: $4