LLM leaderboard
Public rankings of LLM performance. LMSYS Chatbot Arena (1.5M+ human votes), HuggingFace Open LLM Leaderboard, Artificial Analysis (latency + cost + quality). Standard way to compare models.
Chatbot Arena: blind-A/B test between models, users vote, Elo rating. Top 2025: Claude 4.7 Opus, GPT-5, Gemini 2.5 Ultra near-tied. HuggingFace leaderboard: standardized benchmarks (MMLU, GPQA, HellaSwag, etc) on open-weight models. Vellum, Artificial Analysis: track cost + speed + quality for API providers. Modern caveat: leaderboard quality declined — many benchmarks contaminated (in training data) or artificially gamed. Modern best practice: build your own eval suite for domain-specific decisions, don't rely purely on leaderboards.