Fable 5.2 nearly doubles GPT-6 Astra on remote-work automation, but 59% of tasks remain out of reach

Fable 5.2 automated 41% of the tasks in the Remote Labor Index, according to the short note this piece is based on. OpenAI's GPT-6 Astra reached 21% on the same test. That makes the gap roughly two to one.

We only have the headline figures. The source does not say which tasks were included, who ran the evaluation, how many attempts each model got, or what conditions applied. So we read the result as a strong signal, not a final verdict. As the name suggests, the index tries to measure how much remote work a model can actually complete. That is a different question from how well it answers exam-style questions.

The 41% matters more than the ranking. Saturated benchmarks, where most models score close to 100%, can no longer tell models apart. On this test, even the leader fails at most of the work. A low ceiling like that makes the gap between models easier to trust. It is also consistent with the hierarchy we track in the Zendoric Quality index: Anthropic first, OpenAI close behind. Here, though, the gap is much wider than on other benchmarks.

Our reading: this is the kind of number that should drive the jobs debate, because it measures completed tasks rather than impressions. And it argues for neither panic nor complacency. Four in ten tasks done is a serious exposure for remote, digital and standardised work, the same back-office profile that our sector-by-sector series already flagged as the most vulnerable. Six in ten tasks unsolved shows how much of a job is still judgement, context and ending up with a usable result. Short term, expect pressure on outsourced and freelance digital work, and on how that work is priced. Long term, the direction is the one we have argued for: as automation takes on routine work, the people who supervise agents and judge their output gain value. That is the transition to a world where people work on what they care about. Before treating this as a milestone, we need to see the methodology and an independent replication. The figure is useful only if it holds up under outside checks.