We’re proud to announce the launch of AutumnBench, an open-source benchmark developed on our Autumn platform. This benchmark, led by our MARA team, provides a novel platform for evaluating world modeling and causal reasoning in both human and artificial intelligence.