Chatbot Arena ELO rating for instruction following — human-judged precision of following complex requests.
Human ratings capture nuanced failures in intent following that rigid template checks can miss; live prompt distribution improves realism compared with synthetic IF sets.
Per-model sample size varies, so newer models can have unstable ratings; human judges may under-detect subtle constraint violations in long responses. Data: Arena (lmarena.ai) leaderboard dataset, https://huggingface.co/datasets/lmarena-ai/leaderboard-dataset, licensed CC-BY-4.0; rows are converted to the ELO fields shown here.
Higher is better. ELO-like rating for instruction following.
Per-model sample size varies, so newer models can have unstable ratings.
No scores available for this benchmark yet.