Large language models (LLMs) have transformed the software engineering landscape. Recently, numerous LLM-based agents have been developed to address real-world software issue fixing tasks. Despite their state-of-the-art performance, Despite achieving state-of-the-art performance, these agents face a significant challenge: Insufficient high-quality issue descriptions. Real-world datasets often exhibit misalignments between issue descriptions and their corresponding solutions, introducing noise and ambiguity that mislead automated agents and limit their problem-solving effectiveness. We propose SWE-Fuse, an issue-description-aware training framework that fuses issue-description-guided and issue-free samples for training SWE agents. It consists of two key modules: (1) An issue-free-driven trajectory learning module for mitigating potentially misleading issue descriptions while enabling the model to learn step-by-step debugging processes; and (2) An entropy-aware RLVR training module, which adaptively adjusts training dynamics through entropy-driven clipping. It applies relaxed clipping under high entropy to encourage exploration, and stricter clipping under low entropy to ensure training stability. We evaluate SWE-Fuse on the widely studied SWE-bench Verified benchmark shows to demonstrate its effectiveness in solving real-world software problems. Specifically, SWE-Fuse outperforms the best 8B and 32B baselines by 43.0% and 60.2% in solve rate, respectively. Furthermore, integrating SWE-Fuse with test-time scaling (TTS) enables further performance improvements, achieving solve rates of 49.8% and 65.2% under TTS@8 for the 8B and 32B models, respectively.
Figures & tables
Figure 1: Results on SWE-bench Veriefied. SWE-Fuse ranks first among 8B and 32B models.
Figure 2: The case in R2E-Gym Jain et al. (2025) demonstrating an issue and gold-patch mismatch.
Figure 3: The overview of SWE-Fuse.
Project
Total
SubSet
astropy/astropy
22
9
django/django
231
92
matplotlib/matplotlib
34
14
mwaskom/seaborn
2
1
pallets/flask
1
1
psf/requests
8
3
Table 1: Statistics of Project Distribution in SWE-bench Verified and Our Subset.
Statistic
Total
Mean
Min
Max
Valid Trajectories
14,350
-
-
-
Instances
14,329
-
-
-
Projects
111
-
-
-
Interaction Rounds
401,958
28.05
10
98
Token Consumption
281,938,584
19,676.08
4,136
65,115
Table 2: Summary Statistics of the Trajectories Dataset
Model
Scaffold
Training
Resolve Rate (%) ↑
Open-Source Models
Parameters ≈ 7B
Qwen3-8B ( Yang et al., 2025a )
OpenHands
-
7.6
SWE-Gym-7B ( Pan et al., 2025 )
OpenHands
SFT
10.6
SWE-agent-LM-7B ( Yang et al., 2025b )
SWE-agent
SFT
15.2
Lingma-SWE-GPT-7B ( Ma et al., 2025 )
SWESynInfer
SFT
18.2
Table 3: Performance comparison of SWE-Fuse and baselines on the SWE-bench Verified.
Size
Resolve Rate(%) ↑
Resolved Issues ↑
0
13.5
27/200
1k
24.5 ( ↑ +11.0%)
49/200
2k
31.5 ( ↑ +18.0%)
63/200
4k
33.5 ( ↑ +20.0%)
67/200
8k
35.0 ( ↑ +21.5%)
70/200
All
39.0 ( ↑ +25.5%)
78/200
Table 4: Impact of trajectories size on SWE-Fuse performance.
Ratio(%)
Resolve Rate(%) ↑
Resolved Issues ↑
0
33.5
67/200
25
34.0 ( ↑ +0.5%)
68/200
50
34.0 ( ↑ +0.5%)
68/200
75
30.5 ( ↓ -3.0%)
61/200
100
30.5 ( ↓ -3.0%)
61/200
Table 5: Effect of issue descriptions and issue-free samples ratio on performance.
Figure 7: The curve of training reward score of ERLVR in SWE-Fuse.
Figure 8: An example case (astropy-13236) from SWE-bench-Verified resolved by SWE-Fuse.
LLM-driven software engineering agents have become a central testbed for real-world language-model capability, yet their training remains limited by the availability of high-quality SWE tasks. Existing synthetic data methods typically create tasks through fixed mutation or bug-injection procedures, making the resulting distributions largely independent of the agent's own weaknesses and training progress. We introduce Socratic-SWE, a closed-loop self-evolution framework that reuses the agent's historical solving traces as a source of training signal. Rather than treating traces only as evidence for reward computation, Socratic-SWE distills them into structured agent skills that summarize recurring failures and effective repair patterns. These skills then guide the generation of targeted repair tasks in real repositories. Candidate tasks are checked through execution-based validation and scored with a solver-gradient alignment reward, so that the retained tasks are both verifiable and useful for improving the Solver. The updated Solver produces new traces, enabling the task curriculum to adapt over successive rounds. Across SWE-bench Verified, SWE-bench Lite, SWE-bench Pro, and Terminal-Bench 2.0, Socratic-SWE consistently improves over self-evolving baselines under the same compute budget, reaching 50.40% on SWE-bench Verified after three iterations. These results suggest that solving traces can serve as a scalable substrate for self-evolving SWE agents.
Chuan Xiao, Zhengbo Jiao, Shaobo Wang +5
2Shanghai Jiao Tong University · 1AI Data, Alibaba Group
Large language models (LLMs) exhibit strong performance on self-contained programming tasks. However, they still struggle with repository-level software engineering (SWE), which demands (1) deep codebase navigation with effective context management for accurate localization, and (2) systematic approaches for iterative, test-driven code modification to resolve issues. To address these challenges, we propose SWE-Adept, an LLM-based two-agent framework where a localization agent identifies issue-relevant code locations and a resolution agent implements the corresponding fixes. For issue localization, we introduce agent-directed depth-first search that selectively traverses code dependencies. This minimizes issue-irrelevant content in the agent's context window and improves localization accuracy. For issue resolution, we employ adaptive planning and structured problem solving. We equip the agent with specialized tools for progress tracking and Git-based version control. These tools interface with a shared working memory that stores code-state checkpoints indexed by execution steps, facilitating precise checkpoint retrieval. This design enables reliable agent-driven version-control operations for systematic issue resolution, including branching to explore alternative solutions and reverting failed edits. Experiments on SWE-Bench Lite and SWE-Bench Pro demonstrate that SWE-Adept consistently outperforms prior approaches in both issue localization and resolution, improving the end-to-end resolve rate by up to 4.3%.
Kang He, Kaushik Roy
Electrical and Computer Engineering, Purdue University
We introduce SWE-ZERO to SWE-HERO, a two-stage SFT recipe that achieves state-of-the-art results on SWE-bench by distilling open-weight frontier LLMs. Our pipeline replaces resource-heavy dependencies with an evolutionary refinement strategy: (1) SWE-ZERO utilizes large-scale, execution-free trajectories to master code semantics and repository-level reasoning, and (2) SWE-HERO applies targeted, execution-backed refinement to transition these semantic intuitions into rigorous engineering workflows. Our empirical results set a new benchmark for open-source models of comparable size. We release a dataset of 300k SWE-ZERO and 13k SWE-HERO trajectories distilled from Qwen3-Coder-480B, alongside a suite of agents based on the Qwen2.5-Coder series. Notably, SWE-HERO-32B achieves a 62.2% resolution rate on SWE-bench Verified. Furthermore, despite being trained exclusively on Python, our agents demonstrate robust zero-shot transferability on SWE-bench Multilingual, reaching 44.1% and confirming the paradigm's generalizability across diverse languages.
Nikolai Ludwig, Wasi Uddin Ahmad, Somshubra Majumdar +1