Claude Fable 5.2 Completes 41% of Real Freelance Jobs

Human Verified 0

Share to: , and

A freelance desk at dusk with design work on a laptop, audio waveforms on a monitor and an architectural model, illustrating Claude Fable 5.2's result on the Remote Labor Index

Anthropic’s Claude Fable 5.2 completed 41.3 percent of real freelance projects at professional quality on the Remote Labor Index, the benchmark’s operators said — up from 2.5 percent for the best system a year ago, and more than double the next-best model today.

The scoreboard

Scale Labs and the Center for AI Safety run the index, which hands AI agents complete freelance jobs and has human experts grade the results. Fable 5.2 leads at 41.3 percent. Anthropic’s own Opus 5.5 is second at 21.25 percent, ahead of OpenAI’s GPT-6 Astra at 20.83 percent.

Google’s best entry, Gemini Argon 4, sits fifth at 17.08 percent, while older models such as GPT-5.2 and Gemini 3 Pro score below 3 percent. The gap between first and second is nearly twenty points — wider than the gap between second and tenth.

The pace is the story. When the index launched in October 2025, the best system, Manus, managed 2.5 percent. Fable 5 reached about 16 percent in July 2026, Fable 5.1 about 18 percent, and Fable 5.2 now stands at 41.3 percent — sixteen times the launch figure.

How the test works

The index is built from 240 completed projects bought from 358 verified professionals on Upwork, worth about $144,000 in total. Each job averages $632.60 and just under 29 hours of human work, spanning video editing, CAD, game development and data analysis.

Agents get the brief and input files, plus a virtual workspace with professional software such as Blender, GIMP and Audacity. Trained human experts then grade each deliverable against the freelancer’s own work, judging whether a reasonable client would accept the work.

Where it gained — and where it stops

Fable 5.2 gained most in audio and music editing, architecture and interior design, and 3D modeling, Scale Labs said. Nearly fifty of its successful jobs had never been solved by any agent before — but industrial product design and animation remain weak spots.

The figure has limits. It covers 240 curated projects, not the freelance economy, and excludes work needing client contact, physical labor or long evaluation. Scale Labs publishes no confidence interval for the rate, and Fable 5.2 still fell short on 58.7 percent of the jobs.

Our Read

This benchmark measures agents against paid work rather than exam questions, and the jump matters more than the number. A model that clears four in ten real jobs changes what a studio can hand to software — and what it must still keep for people.

The doubling pace is the sharper signal. The Center for AI Safety says the automation rate has roughly doubled every three months, and Fable 5.2 more than doubled its predecessor in four months. If that pace holds, the majority of these jobs falls within the next year or two.

Why It Matters

Freelance platforms are where AI’s labor effect shows up first, and this index measures exactly that work. Australia’s central bank found tentative signs of weaker outcomes for young workers in AI-exposed jobs this week — the same pressure, measured from the demand side.

For Anthropic, the timing is useful. The company prices its market debut this week, and a flagship model that does real client work at twice the rate of its nearest rival is the strongest evidence yet that its technology, not just its valuation, leads the field.

Background

Anthropic released Fable 5.2 on November 9, alongside Mythos 5.2, cutting prices and leading rival benchmarks at launch. The Remote Labor Index result landed on the leaderboard the same day, and was reported more widely on November 10 — days before Anthropic’s market debut.

Source: The Decoder’s report and the Remote Labor Index leaderboard.

Share to: , and

Faheem R. Khattak

Senior automation engineer and AI marketing journo covering LLMs, algorithms, analytics, core updates, Bing search, and technical SEO.

0 Comments

Leave a Reply

Optional Take Tag:

🛡️ Spam protection: Honeypot + Time-Trap Active

This site uses Akismet to reduce spam. Learn how your comment data is processed.