Arena AI Model ELO History (mayerwin.github.io)

Arena AI has introduced a comprehensive chart tracking the performance history of major AI models based on ELO ratings, designed to illuminate the often-hidden trends in AI model updates. This initiative addresses concerns over "nerfing"—instances where updates lead to aggressive censorship, excessive quantization, or degradation of performance. By utilizing API endpoints for testing, Arena sidesteps the discrepancies that frequently arise between raw models and consumer-facing interfaces, which can implement various filters and optimizations that obscure true performance metrics. The chart is built on constantly updated data sourced from the Hugging Face LM Arena Leaderboard, relying on thousands of blind, crowdsourced evaluations to provide a robust view of each model’s capabilities. This development is significant for the AI/ML community because it enhances transparency around how models evolve over time and the real impacts of updates. By tracking the highest-rated model lineage of each AI lab, Arena helps users and researchers identify when a model’s capabilities genuinely decline versus when changes are simply surface-level adjustments. Additionally, the chart clearly marks new releases and visualizes degradation trends, offering a valuable resource for developers and researchers who need a detailed understanding of the performance dynamics of leading AI systems.