# Improving Large Language Models for Code through Runtime Program-State Reasoning

arXiv: https://arxiv.org/abs/2609.34359 · HTML: https://mlsanity.com/paper/2609.34359/

## Key facts

- No citing papers in the map yet (data through 14 Oct 2026).
- Placed in Reasoning for agents, planning and code, one of 17 directions in LLM reasoning.
- Builds on 23 of its 36 references that are in the map.
- First posted on arXiv 28 Sep 2026.

## Placed in

- LLM reasoning / [Reasoning for agents, planning and code](https://mlsanity.com/lr/reasoning-for-agents-planning-and-code/)

## Builds on (most cited first, 23 of 23)

- [SWE-bench: Can Language Models Resolve Real-World GitHub Issues?](https://mlsanity.com/paper/2310.06770/) (2023)
- [LiveCodeBench: Holistic and Contamination Free Evaluation of Large Language Models for Code](https://mlsanity.com/paper/2403.07974/) (2024)
- [Competition-Level Code Generation with AlphaCode](https://mlsanity.com/paper/2203.07814/) (2022)
- [Absolute Zero: Reinforced Self-play Reasoning with Zero Data](https://mlsanity.com/paper/2505.03335/) (2025)
- [Training Software Engineering Agents and Verifiers with SWE-Gym](https://mlsanity.com/paper/2412.21139/) (2024)
- [SWE-smith: Scaling Data for Software Engineering Agents](https://mlsanity.com/paper/2504.21798/) (2025)
- [SWE-Bench Pro: Can AI Agents Solve Long-Horizon Software Engineering Tasks?](https://mlsanity.com/paper/2509.16941/) (2025)
- [SWE-RL: Advancing LLM Reasoning via Reinforcement Learning on Open Software Evolution](https://mlsanity.com/paper/2502.18449/) (2025)
- [TACO: Topics in Algorithmic COde generation dataset](https://mlsanity.com/paper/2312.14852/) (2023)
- [R2E-Gym: Procedural Environments and Hybrid Verifiers for Scaling Open-Weights SWE Agents](https://mlsanity.com/paper/2504.07164/) (2025)
- [SWE-rebench: An Automated Pipeline for Task Collection and Decontaminated Evaluation of Software Engineering Agents](https://mlsanity.com/paper/2505.20411/) (2025)
- [SWT-Bench: Testing and Validating Real-World Bug-Fixes with Code Agents](https://mlsanity.com/paper/2406.12952/) (2024)
- [CyberGym: Evaluating AI Agents' Real-World Cybersecurity Capabilities at Scale](https://mlsanity.com/paper/2506.02548/) (2025)
- [NExT: Teaching Large Language Models to Reason about Code Execution](https://mlsanity.com/paper/2404.14662/) (2024)
- [CodeI/O: Condensing Reasoning Patterns via Code Input-Output Prediction](https://mlsanity.com/paper/2502.07316/) (2025)
- [Kimi-Dev: Agentless Training as Skill Prior for SWE-Agents](https://mlsanity.com/paper/2509.23045/) (2025)
- [CodeRL+: Improving Code Generation via Reinforcement with Execution Semantics Alignment](https://mlsanity.com/paper/2510.18471/) (2025)
- [ExecVerify: White-Box RL with Verifiable Stepwise Rewards for Code Execution Reasoning](https://mlsanity.com/paper/2603.11226/) (2026)
- [Code Execution as Grounded Supervision for LLM Reasoning](https://mlsanity.com/paper/2506.10343/) (2025)
- [Chain of Execution Supervision Promotes General Reasoning in Large Language Models](https://mlsanity.com/paper/2510.23629/) (2025)
- [Self-Execution Simulation Improves Coding Models](https://mlsanity.com/paper/2604.03253/) (2026)
- [StepCodeReasoner: Aligning Code Reasoning with Stepwise Execution Traces via Reinforcement Learning](https://mlsanity.com/paper/2605.11922/) (2026)
- [Enhancing the Code Reasoning Capabilities of LLMs via Consistency-based Reinforcement Learning](https://mlsanity.com/paper/2605.17958/) (2026)

## Abstract (arXiv)

Large language models receive limited explicit training in reasoning about runtime program states. We study whether training models to reason about runtime program states improves downstream software-engineering capabilities. We introduce two complementary program-state reasoning tasks. Buggy input-output reasoning requires a model to generate a concrete input that exposes a behavioral difference between a buggy program and a hidden correct implementation and to predict the resulting execution behavior. Precondition-postcondition reasoning requires an agent to symbolically characterize a bug-triggering precondition, predict the expected postcondition, explain their causal connection, and instantiate this reasoning as an executable regression test. By incorporating these two tasks into a staged post-training pipeline, we develop Comet-9B, a 9B language model based on Qwen3.5-9B Base. We evaluate the resulting checkpoints on repository-level patch generation, regression-test generation, and security PoC generation. Adding both program-state reasoning tasks to supervised fine-tuning (SFT) on issue resolution improves success rates by 7.25 percentage points on SWE-bench Pro and 9.70 points on SWT-Bench Verified. Sequential reinforcement learning on the two tasks yields further gains of 7.25, 26.79, and 4.67 percentage points on SWE-bench Pro, SWT-Bench Verified, and CyberGym, respectively. Despite having only 9B parameters, Comet-9B achieves a score comparable to the reported GPT-5.2 result on SWE-bench Pro and matches the reported success rate of a GPT-4o-based agent on SWT-Bench Verified.