Anthropic announced that its Fable 5.2 model finished 41.3 percent of a benchmark of real freelance jobs. The test, run by Scale Labs, measures whether an AI can deliver professional‑quality results on complete assignments, not just isolated tasks. By comparison, the best system a year ago, Manus, managed only 2.5 percent. Fable 5.2 therefore solved more than sixteen times as many jobs as the early leader. The benchmark covers 240 projects drawn from 358 verified Upwork freelancers. The jobs total about $144,000 in value, with an average project cost of $632.60 and roughly 29 hours of human work. Tasks range from video editing and CAD design to game development and data analysis. Agents work in a virtual workspace that includes professional tools such as Blender, GIMP and Audacity, plus specialized AI generators for image, audio and video. Scale Labs reported that Fable 5.2 made its biggest gains in audio and music editing, architectural and interior design, and 3D model creation. It also handled longer, more complex projects that earlier agents could not finish. Nearly 50 of its successful completions were tasks no AI in the test had solved before. The next‑best models were Anthropic’s Opus 5.5 at 21.25 percent and OpenAI’s GPT‑6 Astra at 20.83 percent. Google’s Gemini Argon 4 placed fifth with 17.08 percent. Older models such as GPT‑5.2 and Gemini 3 Pro fell below three percent. Human experts manually reviewed each AI output by opening the files in the relevant software and comparing them with the human freelancer’s work. Attempts to automate this review with another AI have failed, as reviewers struggle to detect subtle technical flaws.
Why the jump matters
For businesses that hire freelancers, the results suggest that AI agents like Fable 5.2 could start handling a sizable share of routine design and media projects, potentially lowering cost and turnaround time. However, the test also showed weaknesses in industrial product design and animation, and reviewers noted that AI outputs can still be incomplete or technically flawed. Until those gaps are closed, human oversight will remain necessary.
Why it matters
Companies can consider AI agents for a broader set of freelance tasks, especially in audio editing, architecture and 3D modeling, which may cut costs. Yet the lingering errors mean freelancers will still need to review and polish AI output, so the shift will be gradual.
