Anthropic’s Claude Fable 5.2 has an automation rate of 41.30% on the Remote Labor Index, a benchmark of paid freelance projects, according to the RLI leaderboard run by Scale AI and the benchmark’s own site with the Center for AI Safety. That is the share of projects where the model’s work was judged at least as good as the human professional’s. When the leaderboard launched in October 2025, the best agent, Manus, scored 2.5%: Fable 5.2’s score is about 16.5 times that, by our arithmetic.
The next entries are far behind. Anthropic’s Opus 5.5 scores 21.25% and OpenAI’s GPT-6 Astra 20.83%, so Fable 5.2 roughly doubles both, by our arithmetic. Anthropic released the model on 9 November and reports the same 41.3%, at High effort, in its launch benchmark table.
Business Pill · The Model and the Harness
A short explainer on why products built on the same model can perform differently. It teaches the general idea and does not discuss the Remote Labor Index.
The key insight: As we read it, the Remote Labor Index shows the best score on paid freelance projects moving from 2.5% at the October 2025 launch to 41.30% with Anthropic’s Claude Fable 5.2, about 16.5 times higher by our arithmetic, while the next entries on Scale’s leaderboard, Opus 5.5 and GPT 6 Astra, sit at 21.25 and 20.83.
RLI has 240 projects collected from 358 verified freelancers on Upwork, across 23 Upwork domains. Scale says 230 private projects are used for all leaderboard scoring and 10 are public. The projects include game development, productdesign, architecture, data analysis and video animation, and each comes with the brief, the input files and the deliverable the professional actually handed over.
The money is real. Scale lists a total economic value of $143,991, an average projectvalue of $632.60 (median $200) and a mean human completion time of 28.9 hours (median 11.5). The site says costs reach over $10,000 and completion times exceed 100 hours on some projects, all taken from the professionals who did the work.
The test is strict. Trained experts compare each AI deliverable with the human one and ask whether it would be accepted by a “reasonable client”. The automation rate counts projects scored as meeting or exceeding that standard. Scale reports 94.4% inter-annotator agreement using a majority vote of three evaluators.
Remote Labor Index automation rate of selected frontier models, dated by release as on the benchmark site (remotelabor.ai): Opus 4.5 3.8%, GPT-5.5 6.3%, Fable 5 15.8%, GPT-6 Astra 20.8%, Fable 5.2 41.3%. The first point is the 2.5% top score (Manus) Scale reports for the leaderboard launch, dated at the paper’s 30 October 2025 arXiv posting. 16.5x is our arithmetic.
The Leaderboard Now
On Scale’s leaderboard, marked NEW, Fable 5.2 leads at 41.30. Below it: Opus 5.5 at 21.25, GPT 6 Astra at 20.83, Fable 5.1 at 17.92, Gemini Argon 4 at 17.08, Fable 5 at 15.83, Opus 4.8 at 8.33 and Codex GPT 5.5 at 6.25. Anthropic models hold four of the top six places, by our count.
The launch-era agents sit at the bottom. Manus 1.5 and Manus 1.0 are listed at 2.50, ChatGPT agent at 1.25 and gemini-2.5-pro-preview at 0.83. Scale’s page says that “At the time this leaderboard was launched, the highest-performing agent (Manus) achieved a 2.5% automation rate”. The paper behind the benchmark appeared on arXiv on 30 October 2025.
Anthropic’s own table puts the Remote Labor Index under knowledge work and lists Fable 5.1 at 17.9%, Opus 5.5 at 21.3% and GPT-6 Astra at 20.8%. It says all scores show each model’s highest-available effort unless noted; the RLI figure for Fable 5.2 is marked High effort. The benchmark site summarises the trend as an automation pace of “ร2 every ~3.3 months”.
The top eight entries on Scale AI’s Remote Labor Index leaderboard, read on 11 November 2026, as automation rates (%): Fable 5.2 41.30, Opus 5.5 21.25, GPT 6 Astra 20.83, Fable 5.1 17.92, Gemini Argon 4 17.08, Fable 5 15.83, Opus 4.8 8.33, Codex GPT 5.5 6.25. Names as Scale lists them.
Our Numbers
By our arithmetic, 41.30% is 16.5 times the 2.5% launch-time top score, over about 12 months between the paper’s 30 October 2025 posting and Fable 5.2’s 9 November 2026 release. That is about four doublings, or one every three months or so, close to the site’s own ~3.3-month figure.
The step from Anthropic’s previous generation is also large. Fable 5, dated 9 June 2026 on the benchmark site, scores 15.83%; five months later Fable 5.2 is 2.6 times higher, by our arithmetic. Against the best non-Anthropic entries, Fable 5.2 is 1.98 times GPT-6 Astra and 2.42 times Gemini Argon 4.
The other side of 41.30% is 58.70%: on most projects the model’s work still did not meet the professional’s standard, by the benchmark’s own measure. Scale defines a Dollars Earned metric, the original human cost of the projects an agent completes, but the rankings we read report the automation rate, so we do not estimate a dollar figure; project values run from a $200 median to over $10,000.
As we read the leaderboard, the widest gap is inside Anthropic’s own range: Fable 5.2 is 20.05 points ahead of Opus 5.5, by our arithmetic, the model Anthropic’s launch post calls its strongest daily driver and recommends users start with for most coding and knowledge tasks.
As we read the method, the automation rate is pass or fail per project: a deliverable either would be accepted by a reasonable client or it counts as a failure, however close it came. Scale’s separate Elo score exists to capture that partial progress, with the human baseline fixed at 1,000.
Remote Labor Index site, Center for AI Safety and Scale AI
“These results help ground discussions of AI automation in empirical evidence, setting a common basis for tracking progress and enabling stakeholders to proactively navigate AI-driven labor automation”
Three Implications
FOR BUYERS OF FREELANCE WORK Scale lists a median projectvalue of $200 and a mean of 28.9 hours of human work across the benchmark’s projects.
FOR THE LABS Anthropic models hold four of the top six places on Scale’s RLI leaderboard, by our count, with OpenAI and Google entries at 20.83 and 17.08.
FOR MEASUREMENT Scale says RLI excludes projects that need direct client interaction, physical labor or long-term evaluation, so the rate is not a share of all remote work.
The framework argues: “Model capability now compounds on a six-to-twelve-month doubling cycle.”
As we read it, the benchmark’s own site puts the doubling on this test at about 3.3 months, faster than that range. The framework predates the 41.30% result and does not discuss the Remote Labor Index.
What Is Not Established
The pages we read do not say whether Scale and CAIS ran Fable 5.2 themselves or listed a result supplied by Anthropic, or which agent environment it used. At launch, Scale’s method gave every agent a standardized prompt, common tools and a maximum generation budget of $30 per task; Anthropic’s figure is at High effort.
An automation rate on RLI is not a share of all remote work. Scale says RLI “does not represent the entire digital economy”: it excludes projects that need direct client interaction, physical labor or long-term evaluation. Scale also flags a contamination risk on the 10 public projects, which it mitigates by scoring on the 230 private ones.
This Week In The AI Supercycle
The Great Convergence of Capital, Power, and Geopolitics
This week’s issue: for two years the AI build-out was priced as a supply story, and last week as a credit story. This week it was priced as a permission story, and the answer arrived as a cost, not a ban.
Anthropic’s Claude Fable 5.2 completed 41.30% of the Remote Labor Index’s paid freelance projects to a client-acceptable standard, per the Scale AI and CAIS leaderboards, about 16.5 times the 2.5% top score at the October 2025 launch and roughly double the next entries, Opus 5.5 at 21.25% and GPT-6 Astra at 20.83%, by our arithmetic.
A note on sourcing. We read the Remote Labor Index’s own site (remotelabor.ai, from the Center for AI Safety and Scale AI) and Scale AI’s RLI leaderboard page on 11 November 2026, Anthropic’s Fable 5.2 launch post of 9 November and the paper’s arXiv record. Scores are as those pages list them; multiples, counts and the doubling estimate are our arithmetic. We first saw the 41% figure in an OpenTools item and worked from the benchmark pages instead. Anthropic, Scale AI and CAIS were not contacted for this piece.
Gennaro Cuofano is a CRO and tech executive who has worked in AI since late 2015, starting with bringing NLP, natural-language generation, voice and chatbot products to the marketing industry. His background is in law and finance: he holds a Master's degree in Law and an International MBA with an emphasis on corporate finance (LUISS Business School and the University of San Diego, 2012), and worked as a financial analyst at a real-estate investment firm in San Diego and as an assistant controller. His work focuses on business model strategy, business engineering and, more broadly, structural analysis: how companies make money, read from their own filings. He created FourWeekMBA and leads research there and at The Business Engineer, his newsletter on AI and business strategy, with over 95,000 subscribers and more than 1,000 published analyses. His writing has also appeared on Entrepreneur (for example, How to Digitize a Family Business in Europe) and HackerNoon. Find him on LinkedIn and Substack. See how we source.
