Environment Duels benchmark scores LLMs as task authors and solvers
Original title: Environment Duels: A Microcosm of LLM Co-Evolution
Across nine frontier models, the best solvers are not the best authors, and smaller models write the hardest environments.
Environment Duels is a dynamic agentic benchmark in which each model writes interactive environments 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 are distinct skills. Head-to-head outcomes are transitive, independent authors converge on similar designs, and a reinforcement-learning study with Qwen3.8-27B finds that hint lift on an author's environments tracks gains on held-out reasoning and agentic benchmarks. Code is at https://github.com/LLM-Duels/EnvDuels.
- Smaller models learn more from others' play and write the most useful hints.
- The learner's hint lift tracks improvement whether environments are its own or another model's.
- Head-to-head outcomes are transitive and produce a stable ranking.
- AI models
- Qwen3.8-27B
From the original article
1 min read at arxiv.orgThrough 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.
By Shibo Jin · arxiv.org
A short excerpt; the article and its images belong to the publisher. Summary, key facts and assessment above are AI-written from the article and can be wrong.