mlsanityData to 7 Oct 2026

R2E-Gym: Procedural Environments and Hybrid Verifiers for Scaling Open-Weights SWE Agents

arXiv 2504.07164 ↗First posted 9 Apr 2025cs.SE cs.CL cs.LGcited by 140

On the map: LLM reasoning → Reasoning for agents, planning and code

also in 1 more
Abstract (arXiv)

Improving open-source models on real-world SWE tasks (solving GITHUB issues) faces two key challenges: 1) scalable curation of execution environments to train these models, and, 2) optimal scaling of test-time compute. We introduce AgentGym, the largest procedurally-curated executable gym environment for training real-world SWE-agents, consisting of more than 8.7K tasks. AgentGym is powered by two main contributions: 1) SYNGEN: a synthetic data curation recipe that enables scalable curation of executable environments using test-generation and back-translation directly from commits, thereby reducing reliance on human-written issues or unit tests. We show that this enables more scalable training leading to pass@1 performance of 34.4% on SWE-Bench Verified benchmark with our 32B model. 2) Hybrid Test-time Scaling: we provide an in-depth analysis of two test-time scaling axes; execution-based and execution-free verifiers, demonstrating that they exhibit complementary strengths and limitations. Test-based verifiers suffer from low distinguishability, while execution-free verifiers are biased and often rely on stylistic features. Surprisingly, we find that while each approach individually saturates around 42-43%, significantly higher gains can be obtained by leveraging their complementary strengths. Overall, our approach achieves 51% on the SWE-Bench Verified benchmark, reflecting a new state-of-the-art for open-weight SWE-agents and for the first time showing competitive performance with proprietary models such as o1, o1-preview and sonnet-3.5-v2 (with tools). We will open-source our environments, models, and agent trajectories.

Key facts. Cited by 140 papers in the map (data through 7 Oct 2026). Placed in Reasoning for agents, planning and code, one of 17 directions in LLM reasoning. Builds on 29 of its 45 references that are in the map. First posted on arXiv 9 Apr 2025.

Builds on

29 of 45 references.

Closest ancestors · 3

Same direction, newest first

  1. 2024LiveCodeBench: Holistic and Contamination Free Evaluation of Large Language Models for Code12 Mar
  2. 2023LEVER: Learning to Verify Language-to-Code Generation with Execution16 Feb
  3. 2022Competition-Level Code Generation with AlphaCode8 Feb

Other references · 26

  1. 2021Training Verifiers to Solve Math Word Problemscited by 9,743
  2. 2022ReAct: Synergizing Reasoning and Acting in Language Modelscited by 8,355
  3. 2021Evaluating Large Language Models Trained on Codecited by 8,053
  4. 2023SWE-bench: Can Language Models Resolve Real-World GitHub Issues?cited by 3,193
  5. 2024SWE-agent: Agent-Computer Interfaces Enable Automated Software Engineeringcited by 1,628
  6. 2024LlamaFactory: Unified Efficient Fine-Tuning of 100+ Language Modelscited by 1,442
  7. 2024OpenHands: An Open Platform for AI Software Developers as Generalist Agentscited by 956
  8. 2021Measuring Coding Challenge Competence With APPScited by 879
  9. 2024BigCodeBench: Benchmarking Code Generation with Diverse Function Calls and Complex Instructionscited by 527
  10. 2024Agentless: Demystifying LLM-based Software Engineering Agentscited by 400
  11. 2022CodeT: Code Generation with Generated Testscited by 369
  12. 2024Training Software Engineering Agents and Verifiers with SWE-Gymcited by 300
14 more
  1. 2024AutoCodeRover: Autonomous Program Improvementcited by 272
  2. 2025SWE-RL: Advancing LLM Reasoning via Reinforcement Learning on Open Software Evolutioncited by 246
  3. 2024On LLMs-Driven Synthetic Data Generation, Curation, and Evaluation: A Surveycited by 240
  4. 2023Self-Alignment with Instruction Backtranslationcited by 149
  5. 2024Code Generation with AlphaCodium: From Prompt Engineering to Flow Engineeringcited by 100
  6. 2024SWE-Bench+: Enhanced Coding Benchmark for LLMscited by 95
  7. 2025SWE-Fixer: Training Open-Source LLMs for Effective and Efficient GitHub Issue Resolutioncited by 87
  8. 2022Coder Reviewer Reranking for Code Generationcited by 83
  9. 2024Lingma SWE-GPT: An Open Development-Process-Centric Language Model for Automated Software Improvementcited by 57
  10. 2024Commit0: Library Generation from Scratchcited by 50
  11. 2023ALGO: Synthesizing Algorithmic Programs with LLM-Generated Oracle Verifierscited by 41
  12. 2024Diversity Empowers Intelligence: Integrating Expertise of Software Engineering Agentscited by 35
  13. 2022Fault-Aware Neural Code Rankerscited by 33
  14. 2024The Counterfeit Conundrum: Can Code Language Models Grasp the Nuances of Their Incorrect Generations?cited by 20

Built on by

Cited by year (2026 to 7 Oct)

Citing papers per year: 2025: 35; 2026: 105.

140 papers.

Sort:

By area:

  1. 2025SWE-smith: Scaling Data for Software Engineering Agentscited by 255
  2. 2025A Survey of Reinforcement Learning for Large Reasoning Modelscited by 175
  3. 2025From LLM Reasoning to Autonomous AI Agents: A Comprehensive Reviewcited by 146
  4. 2025VerlTool: Towards Holistic Agentic Reinforcement Learning with Tool Usecited by 92
  5. 2025CWM: An Open-Weights LLM for Research on Code Generation with World Modelscited by 87
  6. 2025Nemotron 3 Nano: Open, Efficient Mixture-of-Experts Hybrid Mamba-Transformer Model for Agentic Reasoningcited by 85
  7. 2025CyberGym: Evaluating AI Agents' Real-World Cybersecurity Capabilities at Scalecited by 71
  8. 2025SEC-bench: Automated Benchmarking of LLM Agents on Real-World Software Security Taskscited by 69
  9. 2025Stabilizing MoE Reinforcement Learning by Aligning Training and Inference Routerscited by 67
  10. 2026MOPD: Multi-Teacher On-Policy Distillation for Capability Integration in LLM Post-Trainingcited by 60
  11. 2025A Survey of LLM-Driven AI Agent Communication: Protocols, Security Risks, and Defense Countermeasurescited by 54
  12. 2025Nemotron-Cascade: Scaling Cascaded Reinforcement Learning for General-Purpose Reasoning Modelscited by 45
38 more
  1. 2025SkyRL-Agent: Efficient RL Training for Multi-turn LLM Agentcited by 44
  2. 2025The Complexity Trap: Simple Observation Masking Is as Efficient as LLM Summarization for Agent Context Managementcited by 40
  3. 2025Context as a Tool: Context Management for Long-Horizon SWE-Agentscited by 38
  4. 2025Kimi-Dev: Agentless Training as Skill Prior for SWE-Agentscited by 35
  5. 2025Training Long-Context, Multi-Turn Software Engineering Agents with Reinforcement Learningcited by 34
  6. 2026Nemotron 3 Super: Open, Efficient Mixture-of-Experts Hybrid Mamba-Transformer Model for Agentic Reasoningcited by 33
  7. 2026Nemotron 3 Ultra: Open, Efficient Mixture-of-Experts Hybrid Mamba-Transformer Model for Agentic Reasoningcited by 32
  8. 2026SWE-Lego: Pushing the Limits of Supervised Fine-tuning for Software Issue Resolvingcited by 30
  9. 2026FeatureBench: Benchmarking Agentic Coding for Complex Feature Developmentcited by 30
  10. 2025BugPilot: Complex Bug Generation for Efficient Learning of SWE Skillscited by 28
  11. 2025When Agents go Astray: Course-Correcting SWE Agents with PRMscited by 25
  12. 2025GSO: Challenging Software Optimization Tasks for Evaluating SWE-Agentscited by 24
  13. 2025SWE-RM: Execution-free Feedback For Software Engineering Agentscited by 24
  14. 2026Agentic Rubrics as Contextual Verifiers for SWE Agentscited by 22
  15. 2026SWE-World: Building Software Engineering Agents in Docker-Free Environmentscited by 21
  16. 2026CUA-Gym: Scaling Verifiable Training Environments and Tasks for Computer-Use Agentscited by 21
  17. 2026Immersion in the GitHub Universe: Scaling Coding Agents to Masterycited by 20
  18. 2026CLI-Gym: Scalable CLI Task Generation via Agentic Environment Inversioncited by 20
  19. 2026BeyondSWE: Can Current Code Agent Survive Beyond Single-Repo Bug Fixing?cited by 20
  20. 2026SWE-Master: Unleashing the Potential of Software Engineering Agents via Post-Trainingcited by 18
  21. 2026ProRL Agent: Rollout-as-a-Service for RL Training of Multi-Turn LLM Agentscited by 17
  22. 2025Satori-SWE: Evolutionary Test-Time Scaling for Sample-Efficient Software Engineeringcited by 16
  23. 2025SWE-SQL: Illuminating LLM Pathways to Solve User SQL Issues in Real-World Applicationscited by 15
  24. 2026ResearchGym: Evaluating Language Model Agents on Real-World AI Researchcited by 15
  25. 2025SWE-Dev: Evaluating and Training Autonomous Feature-Driven Software Developmentcited by 13
  26. 2025Part II: ROLL Flash -- Accelerating RLVR and Agentic Training with Asynchronycited by 13
  27. 2025Environment Scaling for Interactive Agentic Experience Collection: A Surveycited by 13
  28. 2025Training Versatile Coding Agents in Synthetic Environmentscited by 13
  29. 2026CM2: Reinforcement Learning with Checklist Rewards for Multi-Turn and Multi-Step Agentic Tool Usecited by 13
  30. 2026daVinci-Env: Open SWE Environment Synthesis at Scalecited by 13
  31. 2026AgentLens: Revealing The Lucky Pass Problem in SWE-Agent Evaluationcited by 13
  32. 2026On Randomness in Agentic Evalscited by 12
  33. 2026PivotRL: High Accuracy Agentic Post-Training at Low Compute Costcited by 12
  34. 2026SWE-TRACE: Optimizing Long-Horizon SWE Agents Through Rubric Process Reward Models and Heuristic Test-Time Scalingcited by 12
  35. 2026Open-MOPD: Diagnosing and Fixing Capability Imbalance in Multi-Teacher On-Policy Distillationcited by 12
  36. 2025AutoSDT: Scaling Data-Driven Discovery Tasks Toward Open Co-Scientistscited by 11
  37. 2025WebGen-Agent: Enhancing Interactive Website Generation with Multi-Level Feedback and Step-Level Reinforcement Learningcited by 11
  38. 2026AutoLab: Can Frontier Models Solve Long-Horizon Auto Research and Engineering Tasks?cited by 11

50 most cited of 140 All 140 (JSON)

Similar, not cited

Close in wording, no citation either way

  1. 2026SWE-rebench V2: Language-Agnostic SWE Task Collection at Scale
  2. 2026SWE-Replay: Efficient Test-Time Scaling for Software Engineering Agents
  3. 2026SWE-Universe: Scale Real-World Verifiable Environments to Millions
  4. 2025Skywork-SWE: Unveiling Data Scaling Laws for Software Engineering in LLMs
  5. 2026SWE-Serve: Benchmarking Agentic Engineering For Production Inference Serving
  6. 2025SWE-Dev: Building Software Engineering Agents with Training and Inference Scaling
  7. 2026Hybrid-Gym: Training Coding Agents to Generalize Across Tasks
  8. 2026SWE-Marathon: Can Agents Autonomously Complete Ultra-Long-Horizon Software Work?