Drex 1.5 uses Snake to compare decision ability against Jev
Two agents play Snake from the same state and actions, with Drex 1.5 scoring 58.28 on Decision Index 0.2.1 against Jev 1.13.0's 57.91, making sub-second single-pass decisions at 128k context.
Drex 1.5 turns Snake into a live stress test for decision models. two agents get the same state and actions, so every move becomes a decision you can actually watch. → Drex 1.5: 58.28 on Decision Index 0.2.1 → Jev 1.13.0: 57.91 → 93.4% accuracy across 32k–128k context → sub-second decisions, single pass built by http://Nace.AI, Drex ranks options and picks actions before a heavier model needs to run. the 128k window is the real upgrade: more state stays inside the decision layer without constant summarizing. Snake is just the demo. the real use case is routing agents, tools and workflows before expensive model calls.