Researchers from the Center for AI Safety and Scale AI have released the Remote Labor Index (RLI), a benchmark testing AI agents on 240 real-world freelance jobs across 23 domains.

They find current AI agents have low but steadily improving performance. The best-performing agent (Manus) successfully completed 2.5% of projects, earning $1,720 out of a possible $143,991. However, newer models consistently perform better than older ones, indicating measurable advancement toward automating remote work.