Fable 5.2 automates 41% of remote work tasks, nearly double GPT-6 Astra, but it is one benchmark

Fable 5.2 completed 41% of the tasks in the Remote Labor Index. That benchmark measures how much real remote work an AI system can do from start to finish. OpenAI's GPT-6 Astra reached 21% on the same test, so Fable 5.2 scored almost twice as high. The short note we have gives no methodology, no task breakdown and no independent check.

If the result holds up, it fits a pattern we have tracked for months. The Fable line, which we have so far attributed to Anthropic, leads our quality index. The lead is widest on long, practical, multi-step work, which is closer to real jobs than to trivia tests.

Our reading: 41% does not mean 41% of remote jobs go away. It means a model can now finish a meaningful share of self-contained tasks without help. That pressure falls first on back-office and freelance work that can be split into separate deliverables. Judgment, client relationships and accountability are still hard to automate. In the short term, expect cheaper work and pressure on rates. In the long term, the same capability lowers the cost of producing things in general, and that is the path toward abundance. We will treat the number as a strong signal once someone outside the lab reproduces it.