Anthropic's Claude Fable 5.2 tops Remote Labor Index at 41.3%

Anthropic's Claude Fable 5.2 scored a 41.3% automation rate on the Remote Labor Index, nearly twice the 21.25% recorded for Anthropic's Opus 5.5. OpenAI's GPT 6 Astra ranked third at 20.83%.

That placed Fable 5.2 first among the six systems listed on the Scale Labs leaderboard. Scale Labs and the Center for AI Safety (CAIS) operate the index. The score did not appear in Anthropic's launch materials for Fable 5.2 and Mythos 5.2, released Nov. 9 and covered by MLQ.ai. Anthropic has not commented on the result.

What does the 41.3% automation rate mean?

Expert human graders, not AI judges, judged Fable 5.2's deliverable at least as good as the human freelancer's work on 41.3% of projects. That met the standard a reasonable client would accept. The benchmark asks AI agents to complete real freelance jobs.

Its project set covered:

• 240 real projects from 358 verified Upwork freelancers.
• About $144,000 in total project value.
• An average project value of $632.60.
• Roughly 29 hours of human work per project.
• Video editing, CAD, game development and data analysis assignments.

Agents operated a virtual desktop with tools including Blender, GIMP and Audacity. Leaderboard scores use 230 private projects from that larger set. Scale Labs publishes no confidence interval for the automation rate.

How did the leading score develop?

The RLI paper, published in October 2025, placed Manus first with a 2.5% completion rate before the field moved higher over the following months. Fable 5 led at about 16% in July. Its direct predecessor, Fable 5.1, scored 17.92%.

Model	Developer	Automation rate
Fable 5.2	Anthropic	41.30%
Opus 5.5	Anthropic	21.25%
GPT 6 Astra	OpenAI	20.83%
Fable 5.1	Anthropic	17.92%
Gemini Argon 4	—	17.08%
Fable 5	Anthropic	15.83%

Fable 5.2 more than doubled the second-place score of Opus 5.5 on the same published sample. That gap widened the distance between first and second place, but it does not remove the limits of a fixed sample. The published scores do not estimate what share of remote work AI can automate. The leaderboard measures performance on the selected projects, not the reach of AI agents across freelance work.

Where did agents succeed, and where did they fail?

Scale Labs found the largest gains in audio and music editing, architecture and interior design, and 3D modeling, according to The Decoder. Nearly 50 of Fable 5.2's successful jobs had not been completed by any earlier agent.

Industrial product design and animation remain weak spots. Scale Labs said the AI "can barely handle" those areas. Typical failures included amateur-quality work, incomplete or broken files, and contradictory designs.

The results show uneven progress across the project categories. Audio and design work improved sharply. Product and animation deliverables still fell short of the human reference work on many assignments.

How does the index grade the work?

Trained experts open each submission in the relevant professional software, then compare it directly with the freelancer's delivered work. CAIS has tried to automate grading with AI reviewers, but those attempts have failed so far. The Decoder reported that the reviewers struggled to operate the software and graded too generously. Human experts therefore remain the basis for the published scores.

That method gives trained human judgment a direct role in every leaderboard result. It also makes failures in file handling and design coherence visible in the scores. The project set is fixed. The absence of a published confidence interval limits any precise estimate of Anthropic's lead.

The next competitive question for Anthropic is whether its lead holds beyond this fixed sample, since the ranking covers 230 private projects without a published confidence interval.