Anthropic’s Claude Fable 5.2 has achieved a significant milestone in AI automation, completing 41.3% of real freelance projects to a standard that human evaluators considered acceptable for paying clients. The result comes from the Remote Labor Index, a benchmark developed by Scale Labs and the Center for AI Safety (CAIS) to measure how effectively AI agents perform actual professional assignments.
The score represents a dramatic improvement from October 2025, when the best-performing AI system completed just 2.5% of the projects in the benchmark. Fable 5.2 now leads the leaderboard by a substantial margin, raising fresh questions about how quickly AI agents could reshape freelance work, professional services and the broader labor market.
Fable 5.2 Reaches 41.3% on the Remote Labor Index
The Remote Labor Index evaluates AI systems using real projects originally completed by professional freelancers. Rather than testing isolated questions or short tasks, it measures whether an AI agent can produce a finished deliverable that meets the expectations of a paying customer.
Anthropic’s Fable 5.2 achieved a 41.3% automation rate in the latest evaluation. This means it produced work judged acceptable or better on more than four out of every ten projects.
The improvement is particularly striking compared with earlier results. When the index launched in October 2025, the leading system, Manus, completed only 2.5% of the assignments at the required standard.
The latest results also show Fable 5.2 moving ahead of other leading AI models.
| AI model | Remote Labor Index score |
|---|---|
| Anthropic Claude Fable 5.2 | 41.30% |
| Anthropic Claude Opus 5.5 | 21.25% |
| OpenAI GPT-6 Astra | 20.83% |
| Google Gemini Argon 4 | 17.08% |
| Anthropic Claude Fable 5.1 | 17.92% |
The figures show that Fable 5.2 nearly doubles the score of the next-best system, Opus 5.5. Its performance also represents more than twice the result achieved by its immediate predecessor, Fable 5.1.
The comparison with older models is even more pronounced. Several earlier systems, including GPT-5.2 and Gemini 3 Pro, scored below 3% in the reported evaluation.
What Is the Remote Labor Index?
The Remote Labor Index was developed by Scale Labs and the Center for AI Safety to measure how much real-world remote work AI agents can perform.
Its dataset began with 550 completed projects collected from 358 verified professionals on Upwork, a major freelance marketplace. Researchers reviewed the submissions and selected 240 projects covering 23 categories of professional work.
The assignments represent approximately $144,000 in combined project value. They include work in areas such as:
- Video editing and animation
- Architecture and interior design
- 3D modelling
- Game development
- Data analysis
- Computer-aided design (CAD)
- Audio editing and music production
Each project includes the original assignment brief, the necessary input files and the completed work submitted by a human freelancer.
The AI agents receive the same materials and access to a virtual working environment containing professional software. Depending on the assignment, this can include tools such as Blender, GIMP and Audacity.
The benchmark then evaluates whether the AI-generated deliverable is good enough for a reasonable client to accept.
This approach makes the index different from conventional AI benchmarks, which often measure whether a model can answer a question correctly or solve a narrowly defined problem.
Human Evaluators Check Whether the Work Is Actually Good
One of the most important features of the Remote Labor Index is its reliance on human evaluation.
Trained reviewers open the AI-generated files in the relevant professional software and compare them with the original freelancer’s completed work. They assess whether the output meets the assignment requirements rather than simply looking convincing in a screenshot.
Projects receive one of three scores:
- 1: The work does not meet the required standard.
- 2: The work meets the standard expected by a paying client.
- 3: The work exceeds the expected standard.
The reported automation rate counts projects receiving a score of 2 or 3.
Three evaluators review each deliverable, with the majority decision determining the final result. Scale Labs reported 94.4% agreement among evaluators.
Researchers have also struggled to automate this grading process using AI. Their experiments found that AI reviewers could be overly generous toward submissions that looked polished but contained technical problems.
For example, an attractive image of a building interior does not necessarily mean the underlying 3D model is properly constructed. A reviewer must inspect the actual project file using the appropriate software to identify such shortcomings.
Audio Editing, Architecture and 3D Modelling Show Major Improvements
Fable 5.2’s strongest improvements reportedly appear in tasks involving audio and music editing, architectural design, interior spaces and 3D modelling.
Scale Labs highlighted the model’s ability to handle longer and more complex assignments that earlier AI agents frequently failed to complete.
The results also included nearly 50 successful projects that no previous AI agent in the benchmark had completed successfully.
However, performance remains uneven across different types of work. Industrial product design and animation continue to be difficult areas, according to the researchers.
AI agents can still produce incomplete files, technically flawed results, amateur-quality deliverables or designs containing contradictions. A project that appears complete at first glance may fail when opened and examined in professional software.
This distinction matters because freelance customers generally pay for usable results, not merely convincing demonstrations of an AI system’s capabilities.
How Quickly Has AI Freelance Performance Improved?
The latest result suggests that AI agents are improving rapidly on practical professional assignments.
Fable 5 scored approximately 16% in July 2026, while Fable 5.1 reached 17.92%. Fable 5.2 has now raised the figure to 41.3%.
<escape>AI performance on the Remote Labor Index</escape>
| Benchmark period or model | Completion rate |
|---|---|
| October 2025 leading system | 2.5% |
| July 2026 Fable 5 | Approximately 16% |
| Fable 5.1 | 17.92% |
| November 2026 Fable 5.2 | 41.30% |
The progression suggests that improvements in AI reasoning, tool use and the ability to manage multi-step tasks are translating into better professional outputs.
However, benchmark results do not automatically translate into equivalent productivity gains across the economy. The index measures performance on a selected collection of projects under controlled conditions. It does not establish how often an AI agent can repeat a successful result, how much each attempt costs or how much human supervision is required in a commercial environment.
What Does This Mean for Freelancers?
The findings could have important implications for people earning money through online freelance platforms.
AI agents are becoming more capable of completing assignments that previously required professional software knowledge and considerable working time. This could increase competition for certain types of projects, particularly when clients can obtain acceptable results at a lower cost.
However, the benchmark does not show that AI can replace 41% of freelancers or eliminate 41% of freelance jobs. It measures the share of tested projects that an AI agent completed to an acceptable standard.
Real freelance work also involves activities that the benchmark does not fully capture, including client communication, interpreting vague requirements, negotiating prices, managing revisions and maintaining long-term customer relationships.
For freelancers, the growing capabilities of AI could make it increasingly important to combine professional expertise with AI tools. People who can direct AI agents, review their outputs and deliver reliable final results may be better positioned than those relying entirely on repetitive production work.
The Bigger Picture
Fable 5.2’s performance highlights a shift in AI development from systems that primarily generate text and images toward agents that can operate software and deliver completed projects.
The improvement from 2.5% to 41.3% in roughly one year is particularly significant because it comes from a benchmark based on real professional assignments rather than artificial exercises. It suggests that AI agents are becoming more capable of performing multi-step work across several industries.
At the same time, the remaining 58.7% of projects that the model did not complete to an acceptable standard demonstrates how much work remains. Professional judgment, technical reliability, complex design and consistent execution continue to present challenges.
Looking Ahead
The next important question is whether AI agents can maintain these results across larger and more varied collections of professional projects. Future Remote Labor Index updates will help establish whether Fable 5.2 represents a lasting improvement or whether performance will vary as assignments become more complex and less predictable.
For the freelance industry, the change is worth watching closely. AI agents are not yet capable of handling every professional assignment, but their rapidly improving ability to deliver client-acceptable work could change pricing, competition and the skills customers value. The biggest impact may come as businesses begin combining human professionals with AI agents to complete projects faster and at lower cost.
SEO Title: Anthropic Fable 5.2 Completes 41% of Real Freelance Jobs
Meta Description: Anthropic’s Fable 5.2 scored 41.3% on the Remote Labor Index, up from 2.5% a year ago, showing rapid progress in AI freelance automation.
URL Slug: anthropic-fable-5-2-completes-41-percent-freelance-jobs
Focus Keyphrase: Anthropic Fable 5.2
Secondary Keywords: Remote Labor Index, AI freelance jobs, AI agents, freelance automation, Anthropic AI model, Scale Labs, Center for AI Safety
Get the day’s top stories in your inbox
One concise email. No spam, unsubscribe anytime.
