Logics-SWE-Qwen3.6-27B
Official model release for One to More, More to One: Category-Aware Iterative Expert Training for Software Engineering Agents.
📄 Paper · 🤗 Dataset: Logics-SWE-Env-2.5K · 💻 GitHub: AgenticBigBang
📰 News
- [2026.09.20] Released the RRE–MOPD paper on arXiv.
- [2026.09.18] Released Logics-SWE-Qwen3.6-27B under the Apache-2.0 license.
🔎 Overview
Logics-SWE-Qwen3.6-27B is a 27B-parameter model developed for repository-level software engineering agents. Starting from the Qwen3.6-27B model used in our study, it combines category-aware expert development with multi-teacher on-policy distillation into a single deployment policy.
Repository-level tasks require agents to navigate code, edit files, execute commands, inspect feedback, and iteratively repair their solutions. Our work starts from the category see-saw: aggregate progress during joint RL can conceal opposing changes across task categories. Category splitting alone does not guarantee stronger experts. We therefore develop the experts through RRE before integrating their learned behaviors into one model.
Stage 1: Refresh–Repair–Expand expert development
SWE Labeler organizes training tasks into three repository-domain categories:
- A: service, data, and security.
- B: user-facing applications.
- C: systems, tooling, and runtimes.
All experts start from the same base model. Refresh–Repair–Expand (RRE) alternates executable-reward RL with supervised replay of each expert's own verified successful trajectories. Mastery refreshes track how task success rates change; repair consolidates successful behavior, and expansion revisits additional tasks to identify the next training frontier. Agentic-miniRL provides the shared long-horizon RL recipe.
Stage 2: Single-policy integration with MOPD
The student starts from the common base and generates its own trajectories. Each training task is routed to its corresponding category expert for token-level supervision. Reference-anchored extrapolation augments imitation, using a positive teacher–reference gap gate. The student receives no direct environment-reward term during this pure distillation stage.
Neither expert development nor integration uses an external model to provide solution trajectories or action targets.
Here, integration means on-policy distillation, not arithmetic averaging of expert weights. Inference uses one model; category labels and separate expert models are not required for deployment.
Intended use: research on repository-level issue resolution, coding agents, long-horizon post-training, expert development, and policy distillation.
📊 Experimental Results
Values are mean task resolution (%) ± population standard deviation across three evaluation rounds, with one candidate patch per task per round.
| Model | Pro-618 | SWE-bench Multilingual |
|---|---|---|
| Base | 52.64 ± 0.28 | 56.22 ± 0.68 |
| Pooled RL | 55.50 ± 0.46 | 55.56 ± 0.83 |
| Balanced RL | 55.34 ± 0.92 | 57.00 ± 1.19 |
| Logics-SWE-Qwen3.6-27B | 58.04 ± 0.20 | 59.00 ± 0.47 |
Compared with the base model, Logics-SWE-Qwen3.6-27B improves mean task resolution from 52.64% to 58.04% on Pro-618 (+5.40 percentage points) and from 56.22% to 59.00% on SWE-bench Multilingual (+2.78 percentage points).
| Final model category | Pro-618 | SWE-bench Multilingual |
|---|---|---|
| A | 58.07 ± 1.13 | 51.39 ± 1.13 |
| B | 59.37 ± 1.02 | 64.00 ± 3.27 |
| C | 56.63 ± 1.10 | 61.74 ± 0.27 |
Evaluation scope
- Pro-618 retains 618 of 731 SWE-bench Pro tasks using documented public audit findings. Its A/B/C counts are 221/201/196. These are subset results, not full SWE-bench Pro scores.
- SWE-bench Multilingual includes all 300 tasks. A/B/C contain 72/25/176 tasks; 27 unrouted tasks remain in the Full denominator.
- Scores measure the complete agent setup, not standalone text generation. Pro uses the study's R2E-Gym-based setup; Multilingual uses a SWE-agent setup. Absolute scores across the two scaffolds are not directly comparable.
- Unsuccessful or missing evaluations remain in the fixed task denominator.
🚀 Quickstart
Transformers: text-generation smoke test
import torch
from transformers import AutoModelForCausalLM, AutoTokenizer
model_id = "Logics-MLLM/Logics-SWE-Qwen3.6-27B"
tokenizer = AutoTokenizer.from_pretrained(model_id)
model = AutoModelForCausalLM.from_pretrained(
model_id,
dtype=torch.bfloat16,
device_map="auto",
)
model.eval()
messages = [{
"role": "user",
"content": "A parser crashes on an empty input file. Describe how you would "
"investigate the cause, implement a fix, and test for regressions.",
}]
inputs = tokenizer.apply_chat_template(
messages,
tokenize=True,
add_generation_prompt=True,
return_dict=True,
return_tensors="pt",
).to(model.device)
with torch.inference_mode():
output = model.generate(
**inputs,
max_new_tokens=1024,
do_sample=True,
temperature=1.0,
top_p=0.95,
top_k=20,
)
completion = output[0, inputs["input_ids"].shape[-1]:]
print(tokenizer.decode(completion, skip_special_tokens=True))
The example follows the Transformers chat-template interface. Its short generation budget is for demonstration only. Repository-level agent use additionally requires the matching system prompt, tool schema, action parser, sandbox, and verifier. Model-generated commands should not be executed automatically outside an isolated, permission-controlled environment.
📦 Training Data and Release Scope
The public Logics-SWE-Env-2.5K release provides 2,553 software-engineering task instances and their matching environment-image references. It is a curated subset of the training tasks used in the paper; it does not contain the complete training corpus or agent trajectories.
The AgenticBigBang GitHub repository additionally provides the label taxonomy, SWE Labeler code, anonymous training manifests, benchmark outcomes, aggregate results, and training and evaluation configuration specifications.
📚 Citation
If you use this model, please cite:
@misc{zhao2026onetomoremoretoone,
title = {One to More, More to One: Category-Aware Iterative Expert Training for Software Engineering Agents},
author = {Jie Zhao and Ziyu Jiang and Suhang Zheng and Minghui Shan and Xiaoxiao Xu and Lin Qu},
year = {2026},
eprint = {2609.23377},
archivePrefix = {arXiv},
primaryClass = {cs.SE},
url = {https://arxiv.org/abs/2609.23377}
}
🙏 Acknowledgements
We thank the Qwen, ROLL, R2E-Gym, SWE-agent, SWE-bench, SWE-bench Pro, and SWE-bench Multilingual contributors, and the research communities behind MiniRL, MOPD, and ExOPD.
License
The model weights are released under the Apache License 2.0.