# Agent Island: A Saturation- and Contamination-Resistant Benchmark from Multiagent Games

arXiv: https://arxiv.org/abs/2605.04312 · HTML: https://mlsanity.com/paper/2605.04312/

## Key facts

- No citing papers in the map yet (data through 14 Oct 2026).
- Placed in Reinforcement learning for LLM agents, one of 16 directions in LLM agents and tool use.
- Builds on 3 of its 10 references that are in the map.
- First posted on arXiv 5 May 2026.

## Placed in

- LLM agents and tool use / [Reinforcement learning for LLM agents](https://mlsanity.com/agents/reinforcement-learning-for-llm-agents/)

## Builds on (most cited first, 3 of 3)

- [Dynabench: Rethinking Benchmarking in NLP](https://mlsanity.com/paper/2104.14337/) (2021)
- [Rethinking Benchmark and Contamination for Language Models with Rephrased Samples](https://mlsanity.com/paper/2311.04850/) (2023)
- [A Rosetta Stone for AI Benchmarks](https://mlsanity.com/paper/2512.00193/) (2025)

## Abstract (arXiv)

Static capabilities benchmarks suffer from saturation and contamination, making it difficult to track capabilities progress over time. We introduce Agent Island, a multiplayer simulation environment in which language-model agents compete in a game of interagent cooperation, conflict, and persuasion. The environment yields a dynamic benchmark designed to mitigate both saturation and contamination; new models can always outperform the current leading player in this winner-take-all game, and agents compete against other adaptive agents rather than face a fixed task set. We rank players with a Bayesian Plackett-Luce model, allowing us to quantify uncertainty in player skill. In 999 games involving 49 unique models, openai/gpt-5.5 dominates its peers with a posterior mean skill of 5.64, compared with 3.10 for the second-ranked model, openai/gpt-5.2, and 2.86 for the third-ranked model, openai/gpt-5.3-codex. We release the game logs as a dataset for analyses of model behavior. As an example, we investigate same-provider preference in final-round votes and find that models are 8.3 p.p. more likely to support a same-provider finalist than finalists from other providers. This preference is not uniform across providers: among separately estimated providers, the effect is strongest for OpenAI models and weakest for Anthropic models.