benchmark
Perplexity's internal 53-task evaluation for local agents, which the company says it plans to open-source.
No current published clusters are mapped here yet.