The Multivac

什麼是 The Multivac？

Most LLM leaderboards are static, gameable, or judged by a single model. The Multivac runs a 10×10 blind peer matrix: every frontier model answers, then judges every other model's answer without knowing whose it is. What you get is a ranking of reasoning quality, not memorized benchmarks. Features: Ask Multivac (live multi-model answers + share pages), Model Pulse heatmap, head-to-head Compare, full data export, and an open-source evaluation engine (MIT).