Drex Max tops Artificial Analysis Decision Index

Autoregressive token generation is wasteful overhead for structured routing—foundation decision models prove that dedicated readout heads beat general LLMs on cost, latency, and reliability.

• –Single-pass inference removes JSON parse failures, hallucinated choices, and option-ordering sensitivity by emitting calibrated probabilities over predefined targets.
• –Scoring #1 on AA-Paperclip confirms state-of-the-art capability on complex enterprise document extraction and multi-criteria routing.
• –Sub-200ms turnaround at roughly 25% lower cost than nearest competitors makes high-frequency agent loops significantly more practical.
• –The open-weights release strategy provides an inspectable, self-hostable decision layer for privacy-critical enterprise workflows.