Pro-level AI agents now complete 16% of freelance work

A new report from the Center for AI Safety (CAIS) and Scale Labs has quantified how well AI agents can do professional quality work on commercial freelancing jobs. Using a Remote Labor Index, researchers evaluated the performance of AI agents on 240 real world jobs valued at $144,000 completed by 358 verified freelancers. Over an eight month period, AI agent performance increased from 2.5% to 16.1%, representing a six-fold improvement in less than a year.

Claude Fable 5 performed best at 16.1%, followed by Claude Opus 4.8 (8.3%) and GPT-5.5 (6.3%). Tasks included 3D modeling, CAD, architecture, graphic design, video/animation, audio, data analysis, and web app development.

One key takeaway from this report is that AI agents are being graded 2.5–3x more leniently by AI judges than human experts. Thus, automated benchmark scores are likely overestimating true agent abilities. But even so, the trend suggests that AI agents are improving rapidly, and we may not be too far away from another significant milestone.