LMArena
Public arena where anonymous model pairs are ranked by human votes.
Overview
LMArena presents two anonymous model responses to a prompt and asks the visitor to choose the better one, then derives a ranking from those votes. It solves the problem that every static benchmark eventually leaks into training data and stops measuring anything. It exists because the prompts are chosen by the public and are therefore unknown in advance, which makes contamination much harder - and its cost is that the votes are self-selected rather than controlled.
Technology Graph
How LMArena connects to the rest of the stack.