Tev: Scaling Long-Horizon Agentic Environments with Dense Progress Rewards

Key points

• Long-horizon agents have drawn growing attention for their ability to work autonomously -- from hours to days -- on economically valuable tasks, a capability essential to automated discovery and recursive self-improvement (RSI).1
• Yet open-source models still struggle on long-horizon tasks, largely for lack of both training data and a working recipe, and scaling both is hard: long, complex tasks are difficult to curate and verify, and their rewards are sparse.1
• It is built on a latent-world-first approach that first constructs a latent world: a complete and correct hidden specification of the task domain, workflow, and expected state after each step.1
• Second, we turn this metric into a training recipe: via progress-potential reward shaping (PPRS), we convert progress into a dense per-step reward that improves training efficiency and makes otherwise-intractable tasks trainable.1

Sources (2)

• [1]Tev: Scaling Long-Horizon Agentic Environments with Dense Progress Rewards
  Hugging Face Daily Papers · Oct 13, 12:00 AM

  First, we introduce Tev-Gym, the first long-horizon agent training dataset with verifiable, dense, progressive rewards -- 1,170 verified task environments across 7 domains.

  Long-horizon agents have drawn growing attention for their ability to work autonomously -- from hours to days -- on economically valuable tasks, a capability essential to automated discovery and recursive self-improvement (RSI).

• [2]Tev: Scaling Long-Horizon Agentic Environments with Dense Progress Rewards
  arXiv (AI, ML, NLP, CV, robotics, multi-agent) · Oct 13, 05:59 PM · same content

Extractive summary: sentences quoted from the sources.