Claude Fable 5.2 automates 41.3% of freelance projects on the Remote Labor Index

The Center for AI Safety and Scale Labs say Anthropic’s Claude Fable 5.2 produced work that human reviewers judged good enough to accept on 41.3% of projects in the Remote Labor Index. When the two organisations launched the benchmark about a year ago, the best system fully automated 2.5% of those projects.

CAIS, in an official post, said the automation rate “is doubling every ~3 months” and that “AIs may soon be able to do all remote work faster and more cheaply than humans.” The official Remote Labor Index site gives the recent pace as doubling every ~3.3 months.

When we launched the Remote Labor Index a year ago with @ScaleAILabs , the best AI could fully automate 2.5% of real freelance projects. Fable 5.2 now automates 41.3%.

The automation rate is doubling every ~3 months. AIs may soon be able to do all remote work faster and more cheaply than humans.

— Center for AI Safety (@CAIS) November 10, 2026

What the index measures

The Remote Labor Index tests AI agents on complete freelance projects, rather than isolated skills. Scale Labs says the set has 240 projects, collected from 358 verified Upwork freelancers across 23 of 64 Upwork domains. Ten projects are public. Official leaderboard scores use a private set of 230. The work leaves out projects that need physical labour, long-term evaluation, or direct client interaction.

Scale lists a total value of US$143,991, a mean human completion time of 28.9 hours, and an average project value of US$632.60. The official site says the projects represent more than 6,000 hours of real work, with some costing more than US$10,000 and taking more than 100 hours.

Trained human reviewers compare each agent’s deliverable with the professional’s gold-standard work. The headline metric, the automation rate, is the share of projects where the AI’s work is judged at least as good as the human’s, such that a reasonable client would accept it. Scale reports 94.4% agreement among reviewers on that metric, using a majority vote from three of them. In a July post, CAIS said an automated judge it had built overstated how well newer models did, and that human reviewers were still needed.

How the latest models scored

On the Scale Labs leaderboard, Fable 5.2 is first at 41.3%. Claude Opus 5.5 is second at 21.25%, and OpenAI’s GPT-6 Astra is third at 20.83%. Fable 5.1 stands at 17.92%, Google’s Gemini Argon 4 at 17.08%, and Fable 5 at 15.83%, in line with the 15.8% CAIS published for Fable 5 in July. Anthropic’s Fable 5.2 announcement lists the same 41.3% figure, as a High effort score.

🥇 @AnthropicAI’s Fable 5.2 takes the top spot on our Remote Labor Index with @CAIS, our leaderboard which compares AI agents’ work head-to-head with work by human professionals on real-world remote tasks.

The model successfully automates 41.3% of projects, nearly doubling the performance of the next-best model Opus 5.5, a major leap in AI’s ability to handle real-world work.

A closer look at where the biggest jumps are coming from 🧵

— Scale Labs (@ScaleAILabs) November 9, 2026

In the same thread, Scale Labs says the biggest gains were in audio and music, architecture and interior design, and 3D modelling, and that industrial product design and animation remain far from solved. It also says nearly 50 of Fable 5.2’s successful completions were projects no earlier agent in the test had finished.

We’ve already covered Anthropic’s Fable 5.2 and Mythos 5.2 launch, including early tests of the model.

The index still leaves more than half of its projects unautomated at the current top score, and it doesn’t cover every kind of remote work. CAIS’s remark that AIs may soon do all remote work faster and more cheaply than people is a projection from the recent pace, not a finding from the 41.3% result itself.