Hugging Face now has 48 official benchmarks, mapped in one Space

Hugging Face marks 48 datasets as official benchmarks, each with its own leaderboard. A new Space aggregates them: 17 benchmarks cover agents, but science and knowledge draw the most entries (348); 55% of entries come from China and 23% from the USA.

• 48 official benchmarks: 17 for agents, 6 science, 5 documents and OCR
• About 1,020 entries: 348 science, 197 agents, 158 coding
• Median benchmark has ~12 entries; 14 have five or fewer
• 410 models from 95 organizations; 55% of entries from China, 23% from USA