mlsanityData to 14 Oct 2026

PreciseCoder: Faithful and Generalizable Debugging Beyond Unit-Test Rewards

arXiv 2610.18755 ↗First posted 13 Oct 2026cs.SE cs.AI cs.CLcited by 0

On the map: Deep reinforcement learning methods → RL training of language and agent models

Abstract (arXiv)

Reinforcement learning (RL) from unit tests is the common recipe for post-training code models, but a unit test only asks whether the final program passes. We analyze what this unit-test-only reward teaches a model during debugging and find that, while it raises pass rates, it also teaches the model to edit more and to hallucinate bugs in code that is already correct. We introduce PreciseCoder, a training algorithm that augments the unit-test reward with a signal for semantically verified, localized edits, which correlates with smaller edits and higher pass rates. On 4B, 9B, and 27B models, PreciseCoder reduces the average number of edit lines from 7-15to 3 at a pass rate on par with unit-test-only training. Trained only on single-line bugs, PreciseCoder can generalize to repository-level program repair, where it resolves more issues with smaller patches. Its precision also transfers from weaker to stronger models: given the reasoning trace of a 9B PreciseCoder model, frontier models such as GPT-5.6 Sol and Qwen3.8-Max gain 4.6-12.4 points in edit precision with no additional training. Linguistic analysis of the thinking traces shows that our reward encourages the model to reason about individual lines, name the buggy ones, and form consistent bug hypotheses.

Key facts. No citing papers in the map yet (data through 14 Oct 2026). Placed in RL training of language and agent models, one of 15 directions in Deep reinforcement learning methods. Builds on 43 of its 60 references that are in the map. First posted on arXiv 13 Oct 2026.

Builds on

43 of 60 references.

References · 43

  1. 2024DeepSeekMath: Pushing the Limits of Mathematical Reasoning in Open Language Modelscited by 9,379
  2. 2021Evaluating Large Language Models Trained on Codecited by 8,320
  3. 2023Direct Preference Optimization: Your Language Model is Secretly a Reward Modelcited by 8,065
  4. 2025DeepSeek-R1: Incentivizing Reasoning Capability in LLMs via Reinforcement Learningcited by 6,330
  5. 2023Efficient Memory Management for Large Language Model Serving with PagedAttentioncited by 5,944
  6. 2023SWE-bench: Can Language Models Resolve Real-World GitHub Issues?cited by 3,373
  7. 2024HybridFlow: A Flexible and Efficient RLHF Frameworkcited by 2,057
  8. 2024LiveCodeBench: Holistic and Contamination Free Evaluation of Large Language Models for Codecited by 2,004
  9. 2024SWE-agent: Agent-Computer Interfaces Enable Automated Software Engineeringcited by 1,737
  10. 2022Competition-Level Code Generation with AlphaCodecited by 1,462
  11. 2023Is Your Code Generated by ChatGPT Really Correct? Rigorous Evaluation of Large Language Models for Code Generationcited by 1,391
  12. 2024Tulu 3: Pushing Frontiers in Open Language Model Post-Trainingcited by 1,053
31 more
  1. 2026DeepSeek-V4: Towards Highly Efficient Million-Token Context Intelligencecited by 1,040
  2. 2023Large Language Models Cannot Self-Correct Reasoning Yetcited by 1,017
  3. 2023PyTorch FSDP: Experiences on Scaling Fully Sharded Data Parallelcited by 660
  4. 2024BigCodeBench: Benchmarking Code Generation with Diverse Function Calls and Complex Instructionscited by 546
  5. 2023Weak-to-Strong Generalization: Eliciting Strong Capabilities With Weak Supervisioncited by 467
  6. 2024Agentless: Demystifying LLM-based Software Engineering Agentscited by 421
  7. 2024Training Software Engineering Agents and Verifiers with SWE-Gymcited by 336
  8. 2024CRUXEval: A Benchmark for Code Reasoning, Understanding and Executioncited by 300
  9. 2025SWE-smith: Scaling Data for Software Engineering Agentscited by 285
  10. 2025SWE-RL: Advancing LLM Reasoning via Reinforcement Learning on Open Software Evolutioncited by 263
  11. 2026GDPO: Group reward-Decoupled Normalization Policy Optimization for Multi-reward RL Optimizationcited by 193
  12. 2024LLM Critics Help Catch LLM Bugscited by 147
  13. 2025LocAgent: Graph-Guided LLM Agents for Code Localizationcited by 86
  14. 2024DebugBench: Evaluating Debugging Capability of Large Language Modelscited by 76
  15. 2026DeepSeek-V4.1-Flash: Pushing the Limits of KV Cache Compressioncited by 76
  16. 2025On the Use of Agentic Coding: An Empirical Study of Pull Requests on GitHubcited by 55
  17. 2025Kimi-Dev: Agentless Training as Skill Prior for SWE-Agentscited by 41
  18. 2026On the Design of Qwen3.8-Next Architecture: Evaluation, Efficiency, and Training Stabilitycited by 35
  19. 2024Weak-to-Strong Reasoningcited by 33
  20. 2024NoFunEval: Funny How Code LMs Falter on Requirements Beyond Functional Correctnesscited by 26
  21. 2025How to Train Your Advisor: Steering Black-Box LLMs with Advisor Modelscited by 24
  22. 2025MO-GRPO: Mitigating Reward Hacking of Group Relative Policy Optimization on Multi-Objective Problemscited by 23
  23. 2026Overeager Coding Agents: Measuring Out-of-Scope Actions on Benign Taskscited by 13
  24. 2026Coding Agents Don't Know When to Actcited by 7
  25. 2026PAFT: Preservation Aware Fine-Tuning for Minimal-Edit Program Repaircited by 3
  26. 2026When Models Edit Too Much: On the Fidelity of Minimal Code Editscited by 3
  27. 2025SWE-IF: Aligning Code Evaluation with Human Preferencecited by 2
  28. 2026QiMeng-PRepair: Precise Code Repair via Edit-Aware Reward Optimizationcited by 2
  29. 2026Precise Debugging Benchmark: Is Your Model Debugging or Regenerating?cited by 2
  30. 2026Refine After Generation: Toward Correct and Concise Patches in LLM-based Program Repaircited by 2
  31. 2026Reasoning that Travels: Dissecting How Chain-of-Thought Transfers Across Modelscited by 1

Built on by

Not cited yet.

Similar, not cited

Close in wording, no citation either way

  1. 2026TIPCODER: Reinforcement Learning Boosted Test-time Instruction Proposer for Code Generation
  2. 2026BoostAPR: Boosting Automated Program Repair via Execution-Grounded Reinforcement Learning with Dual Reward Models
  3. 2026TaPR: Test-Aware Policy Refinement for Feedback-Conditioned Code Generation
  4. 2026CVeDRL: An Efficient Code Verifier via Difficulty-aware Reinforcement Learning
  5. 2025CodeReasoner: Enhancing the Code Reasoning Ability with Reinforcement Learning
  6. 2026StepCodeReasoner: Aligning Code Reasoning with Stepwise Execution Traces via Reinforcement Learning
  7. 2026RLPF: Reinforcement Learning from Performance Feedback for Code Generation
  8. 2025ACECODER: Acing Coder RL via Automated Test-Case Synthesis