license, base_model, base_model_relation, pipeline_tag, library_name, tags

license	base_model	base_model_relation	pipeline_tag	library_name	tags
apache-2.0	Qwen/Qwen3-8B	finetune	text-generation	transformers	agent multi-turn reinforcement-learning

Qwen3-8B-UserRL-Agent

The UserRL baseline from MIMESIS: Qwen3-8B trained with multi-turn GRPO following UserRL, with GPT-5.5 as the training user simulator. The main text of the paper calls it GRPO w. GPT-5.5.

Paper · Code · Project page · Collection

Results

Mean task score of the five Qwen3-8B agents from the paper across eight UserRL environments, three of them held out from training (IntentionGym, TelepathyGym, SearchGym), under nine evaluation user models that were not used in training.

Agent	Training user	Objective	Mean score
Base	none	SFT	18.49
UserRL (this model)	GPT-5.5	GRPO	26.10
InfoPO	MIMESIS-9B	InfoPO	26.65
UserRL+	MIMESIS-9B	GRPO	29.54
CSD	MIMESIS-9B	GRPO + CSD	31.09

With the GRPO objective fixed, replacing GPT-5.5 with MIMESIS-9B as the training user (UserRL+) improves the mean score under every evaluation user. Each score is the mean over the nine evaluation users; per-user and per-environment scores are in Appendix D of the paper.

Training

RL starts from the Base checkpoint, Qwen3-8B after a supervised warm start, and trains on five UserRL environments: TravelGym, TurtleGym, FunctionGym, TauGym and PersuadeGym. The full configuration is in Appendix D of the paper, and the training scripts are in the agent directory of the code release.

Usage

The model loads like Qwen3-8B. To evaluate it in the eight environments, install the agent code, serve a user model as described there, and run from the agent directory:

bash scripts/run_eval.sh phanviethoang1512/Qwen3-8B-UserRL-Agent userrl

Citation

@article{phan2026mimesis,
  title   = {{MIMESIS}: Learning User Simulators as Training Environments for Interactive Agents},
  author  = {Phan, Hoang and Huynh, Dat and Zhmoginov, Andrey and Zeng, Qi and Mu, Wancen and Cao, Yue and Bi, Shengjie and He, Yun and Oh, Changdae and Lei, Deren},
  journal = {arXiv preprint arXiv:2610.09484},
  year    = {2026}
}