Rate the Models gives benchmarks human context

Its strongest idea is adding real-world usability signals to increasingly noisy model benchmarks, though its usefulness will depend on participation and rating quality.

• –Human ratings capture writing style, reliability, speed, and day-to-day experience that static benchmarks miss
• –One rating per person per model helps limit spam and repeated voting
• –A single average score can still hide major differences across coding, reasoning, voice, and creative workloads
• –The catalog needs enough active raters and task-specific context to compete with established model leaderboards
• –Reviews and rating distributions could become more valuable than the headline score as the dataset grows