Anthropic's Fable 5.2 completes 41.3% of real freelance jobs in new benchmark
AIScale Labs reports that Anthropic's Fable 5.2 completed 41.3% of jobs in its RLI benchmark, up from about 16% for Fable 5 in July and 2.5% for the top system when the benchmark launched a year earlier. The test uses 240 completed Upwork projects worth about $144,000, judged by human experts, and the source says the agent still struggles with industrial product design and animations.
AIWhy it matters.
The source compares Fable 5.2 with prior systems on 240 paid Upwork projects, showing where agents still fail, which helps readers judge how close agents are to real freelance work.
Source details
The Decoder · Maximilian Schreiner · the-decoder.comPublished · added here
Original title: Anthropic's Fable 5.2 completes 41 percent of real freelance jobs, up from 2.5 percent a year ago