Arena Leaderboard

The canonical place to check which LLM actually wins in blind human preference tests, not which one has the best marketing budget. Chatbot Arena runs real pairwise battles where humans vote on outputs without knowing which model produced them — the leaderboard aggregates those votes into Elo scores across dozens of models. If you are deciding which API to build on top of, this is the benchmark that is hardest to game because it is based on real human choices rather than curated eval sets. The data refreshes as new votes come in, so it tracks model updates faster than most academic benchmarks. Honest reservation: the voting population skews toward English-language technical users, so if your product targets other languages or non-technical audiences, the rankings may not reflect your actual users. Still the most trustworthy single-page answer to the question of which model to call first. -> Best for: AI engineer or solo founder choosing a foundation model for a new product