LMArena

LMArena began in April 2023 as Chatbot Arena, a research project of UC Berkeley's LMSYS Org (Sky Computing Lab), and became the most-cited LLM benchmark by ranking models on real human preference: two anonymous models answer the same prompt side by side, users vote, and a modified Bradley-Terry/Elo system produces the leaderboard. It has since expanded into category-specific arenas for text, vision, video, web development, search, image generation, and repo-grounded coding, with over 400 models tracked and millions of votes cast.

In April 2025 it incorporated as the independent company Arena Intelligence Inc., raising a $100 million seed at a $600 million valuation, followed by a $150 million Series A in January 2026 at a $1.7 billion post-money valuation led by Felicis and UC Investments. Its commercial AI evaluation product, launched September 2025, was already at a $30M+ annualized run rate by December 2025. On January 28, 2026 the company dropped the "LM" prefix and rebranded to Arena (arena.ai).

Limitation: researchers have documented methodology biases — including favoring larger providers and open-source models being disadvantaged — so Elo should be combined with capability benchmarks rather than used alone. Note conversations may be shared with model providers and disclosed publicly for research.

Use the free public platform at arena.ai (formerly lmarena.ai); no account needed to vote or browse leaderboards.