Fable 5.2 Tops ECI, Remote Labor and RoboDojo Benchmarks
The signal here is not the score itself but which benchmarks it came from: end-to-end freelance work and robotics control tests with room left to run, so a above-trend result carries more weight than saturated wins. Teams benchmarking frontier models for real workloads can treat Remote Labor-style agentic evaluation as a better filter than knowledge quizzes. Benchmark saturation itself is the emerging buying-criterion debate worth following.
