Sources
WorldLines: A Benchmark for Long-Horizon, Stateful Embodied Agents
WorldLines (arXiv:2606.18847) introduces a benchmark and models for embodied agents that must carry state across many steps rather than act one-shot, targeting the durability-of-state gap most agent evals skip. It's aimed at realistic, long-duration tasks instead of single-step actions. Relevant to anyone building agents that run for hours, not turns.
↳ Follow the thread