PokéAgent Challenge

The PokéAgent Challenge is an AI benchmark that leverages the complexity of Pokémon to evaluate long-horizon planning and game-theoretic reasoning. The challenge is organized into two tracks: competitive battling and RPG speedrunning. In the Battling Track, agents compete head-to-head in competitive Pokémon matches, while in the Speedrunning Track, agents attempt to complete the full Pokémon RPG as quickly as possible. The PokéAgent Challenge began as a NeurIPS 2025 competition and is now an ongoing benchmark with a live leaderboard.

Join us on Discord Read the Paper

Click below for more information about each track and its leaderboard.

PokéAgent Challenge Team

Seth Karten Princeton Jake Grigsby UT Austin

Stephanie Milani NYU / Johns Hopkins Kiran Vodrahalli Google DeepMind Amy Zhang UT Austin Fei Fang CMU Yuke Zhu UT Austin Chi Jin Princeton

@inproceedings{karten2025pokeagent,
  title        = {The PokeAgent Challenge: Competitive and Long-Context Learning at Scale},
  author       = {Karten, Seth and Grigsby, Jake and Milani, Stephanie and Vodrahalli, Kiran
                  and Zhang, Amy and Fang, Fei and Zhu, Yuke and Jin, Chi},
  booktitle    = {NeurIPS Competition Track},
  year         = {2025},
  month        = apr,
}