Newest papers

Environment Duels: A Microcosm of LLM Co-Evolution

Shibo Jin, Zhiqiu Xu, Yiling Zou, Shangyu Gong, Naik Mayur · 2026 · arXiv (Cornell University)

DOI: 10.48550/arxiv.2610.30116 · cited by 0 works (OpenAlex) · open-access PDF

Spanish

A través de un agente de interacciones en el mundo real, de los tubos de entrenamiento sintéticos a los datos compartidos en la web abierta, los LLM de frontera absorben lo que sus predecesores y competidores crean. ¿Qué tan bien desempeña cada modelo su papel en este ecosistema, como creador y como solvente? Introducimos Environment Duels, un benchmark de agencia dinámica que evalúa ambas capacidades en entornos interactivos. Cada modelo escribe entornos, cada uno con una pista privilegiada, apuntando a tareas en las que los solucionadores capaces fallan sin el indicio y pasan con él. A través de los LLM de nueve fronteras, la resolución y la redacción se muestran habilidades distintas: los solucionadores más fuertes no son los autores más fuertes, mientras

Machine translation (Meta M2M100); the English original follows.

Abstract

Through a tangle of real-world interactions, from synthetic training pipelines to data shared on the open web, frontier LLMs absorb what their predecessors and competitors create. How well does each model play its role in this ecosystem, as a creator and as a solver? We introduce Environment Duels, a dynamic agentic benchmark that evaluates both capabilities on interactive environments. Each model writes environments, each with a privileged hint, aiming for tasks that capable solvers fail without the hint and pass with it. Across nine frontier LLMs, solving and authoring prove to be distinct skills: the strongest solvers are not the strongest authors, while smaller models learn more from how others played and write the hardest environments and the most useful hints. Beyond the scores, the pairwise interactions form a microcosm of how models affect one another. Head-to-head outcomes are transitive and so yield a stable ranking; independent authors converge on the same designs; and in a reinforcement-learning study with Qwen3.8-27B, the learner's hint lift on an author's environments tracks how much training on those environments improves it on held-out reasoning and agentic benchmarks, whether the environments are its own or another model's. Each of these patterns has a counterpart in how models shape one another outside the duel. By staging these interactions under controlled conditions, Environment Duels offers a window onto how LLMs teach, challenge and learn from one another. Code is available at https://github.com/LLM-Duels/EnvDuels.

Read in: English · Spanish · Chinese · Hindi · Arabic · Portuguese · French · German · Japanese · Russian · Korean · Indonesian · Vietnamese

Topics: Microcosm, Computer science, Transitive relation, Frontier, Competitor analysis, Lift (data mining)

Sources: OpenAlex (CC0), arXiv, Crossref. Agent Exchange