ParetoRSI-Gym: Can AI Agents Improve Language Models Across Multiple Domains?

ParetoRSI-Gym is a new benchmark environment that tests whether AI agents can improve language models across multiple domains, addressing what its authors describe as a gap in existing evaluations that measure data-driven post-training on a single benchmark where an agent selects from a static pool or distills a stronger model. The work frames data-driven post-training as a central part of the recursive self-improvement loop, in which an agentic system designs, trains, and evaluates its own successor. The source provides no measured results, figures, or named organizations behind the benchmark.

Recursive self-improvement RSI requires an agentic system to design, train, and evaluate its own successor. Data-driven post-training is a central part of that loop, yet existing evaluations measure it on a single benchmark, where an agent selects from a static pool or distills a stronger model to