Self-HPO-Qwen3.5-35B-A3B

The policy trained by Self-HPO from Qwen3.5-35B-A3B, where the policy itself proposes harness revisions in each H-step. Use it with the self-hpo harness.

Setting	Policy	Harness	SWE-Verified	SWE-Multi	SWE-Pro	Avg.
Base	Qwen3.5-35B-A3B	seed	61.0	54.2	58.3	57.8
Self-HPO	Self-HPO-Qwen3.5-35B-A3B	self-hpo	69.5	59.7	65.2	64.8

Resolve rates (%) on SWE-Bench Verified, SWE-Bench Multilingual, and SWE-Bench Pro.

Usage

bash scripts/eval/serve_model.sh --model self-hpo/Self-HPO-Qwen3.5-35B-A3B --ngpu 8
bash scripts/eval/gen_traj.sh --method Self-HPO --benchmark sbv --harness self-hpo --ngpu 8

📄 Paper: https://arxiv.org/abs/2610.22791

💻 Code: https://github.com/MSR-Orchard/Self-HPO

Citation

@article{wei2026selfhpo,
  title   = {Self-Evolving Coding Agents via Harness-Policy Optimization},
  author  = {Wei, Zhepei and Yao, Wenlin and Peng, Baolin and Wang, Danqing and Wu, Qianhui and Cheng, Hao and Meng, Yu and Gao, Jianfeng},
  journal = {arXiv preprint arXiv:2610.22791},
  year    = {2026},
  url     = {https://arxiv.org/abs/2610.22791}
}