# MarioGPT: Open-Ended Text2Level Generation through Large Language Models

*Source* · `src-sudhakaran-2023-mariogpt` · status: **n/a**

| field | value |
|---|---|
| kind | Preprint |
| authors | Shyam Sudhakaran, Miguel González-Duque, Claire Glanois, Matthias Freiberger, Elias Najarro, Sebastian Risi |
| venue | arXiv |
| date | 2023-02-12 |
| arxiv | 2302.05981 |
| accessed | 2026-10-24 |

## Notes

DistilGPT2 (96M parameters with cross-attention) trained on 37 path-annotated levels; 88.4% of 250 generated levels completed by an A* agent versus about 31% for an LSTM baseline; text prompts for pipes, enemies, blocks, elevation; novelty search variant.

## Backlinks (computed)

**cited by:** [Learned level and map generation (PCGRL, MarioGPT, LLM generators)](../technique/learned-level-generation.md)

## Citation

Shyam Sudhakaran, Miguel González-Duque, Claire Glanois, Matthias Freiberger, Elias Najarro, Sebastian Risi (2023-02-12). *MarioGPT: Open-Ended Text2Level Generation through Large Language Models*. arXiv https://arxiv.org/abs/2302.05981

---
Provenance: created in run `2026-10-24-techniques-rl-design`.