Anthropic's Fable 5 hits 16% automation record but can't replace freelancers yet

Anthropic's Fable 5, recently reinstated by the US government, has achieved a record 16.1% automation rate on the Center for AI Safety's Remote Labor Index (RLI), a benchmark that measures how often AI agents can complete freelance projects at a quality clients would accept. This performance doubles Anthropic's own Opus 4.8 (8.3%) and handily beats OpenAI's GPT-5.5 (6.3%). The RLI tests involve tasks like 3D ring rendering, video ad creation, and floor plan mapping, with human evaluators scoring outputs against professional standards. Even after accounting for premature testing (Fable 5 was shut down mid-June), its worst-case score of 14.6% still topped all prior models.

This rapid acceleration — CAIS notes that AI agent capabilities have "quadrupled in under eight months" — signals fast progress in economically valuable automation. Still, a 16.1% success rate means over 83% of freelance tasks remain out of reach. Organizations would need complex multi-agent systems to fully replace human freelancers, and AI still struggles with evaluating its own work: when CAIS tried replacing human judges with an LLM, it failed, as opening project files in professional software remains a weak spot for agents. For now, AI augments rather than replaces freelancers.