SWE-bench reveals that even top language models like Claude 2 resolve only 1.96% of 2,294 real-world GitHub issues, highlighting a gap in practical coding capabilities.
hub Canonical reference
Is Your Code Generated by ChatGPT Really Correct? Rigorous Evaluation of Large Language Models for Code Generation
Canonical reference. 73% of citing Pith papers cite this work as background.
38
Pith papers citing it
Background
73% of classified citations
abstract
Program synthesis has been long studied with recent approaches focused on directly using the power of Large Language Models (LLMs) to generate code. Programming benchmarks, with curated synthesis problems and test-cases, are used to measure the performance of various LLMs on code synthesis. However, these test-cases can be limited in both quantity and quality for fully assessing the functional correctness of the generated code. Such limitation in the existing benchmarks begs the following question: In the era of LLMs, is the code generated really correct? To answer this, we propose EvalPlus -- a code synthesis evaluation framework to rigorously benchmark the functional correctness of LLM-synthesized code. EvalPlus augments a given evaluation dataset with large amounts of test-cases newly produced by an automatic test input generator, powered by both LLM- and mutation-based strategies. While EvalPlus is general, we extend the test-cases of the popular HumanEval benchmark by 80x to build HumanEval+. Our extensive evaluation across 26 popular LLMs (e.g., GPT-4 and ChatGPT) demonstrates that HumanEval+ is able to catch significant amounts of previously undetected wrong code synthesized by LLMs, reducing the pass@k by up-to 19.3-28.9%. We also surprisingly found that test insufficiency can lead to mis-ranking. For example, both WizardCoder-CodeLlama and Phind-CodeLlama now outperform ChatGPT on HumanEval+, while none of them could on HumanEval. Our work not only indicates that prior popular code synthesis evaluation results do not accurately reflect the true performance of LLMs for code synthesis, but also opens up a new direction to improve such programming benchmarks through automated testing. We have open-sourced our tools, enhanced datasets as well as all LLM-generated code at https://github.com/evalplus/evalplus to facilitate and accelerate future LLM-for-code research.
hub tools
citation-role summary
background 8
dataset 3
citation-polarity summary
representative citing papers
AlgoBench creates traceable variants of competitive programming problems via constraint shifts that invalidate original algorithms, paired with complexity metrics that reveal LLMs often produce functionally correct but asymptotically unsuitable solutions.
Empirical evaluation of three LLMs finds prevalent overconfidence in insecure code generation, with security calibration outperforming functional calibration but both degrading in repository-level settings.
ProgramBench introduces 200 tasks where models must reconstruct full programs like FFmpeg or SQLite from docs alone; none of 9 evaluated LMs fully solve any task and the best passes 95% tests on only 3% of tasks while favoring monolithic code.
ClassEval-Pro benchmark shows frontier LLMs achieve at most 45.6% Pass@1 on class-level code tasks, with logic errors (56%) and dependency errors (38%) as dominant failure modes.
Structurally rich task descriptions make LLMs robust to prompt under-specification, and under-specification can enhance code correctness by disrupting misleading lexical or structural cues.
SWE-agent introduces a custom agent-computer interface that lets LM agents solve software engineering tasks, reaching 12.5% pass@1 on SWE-bench and 87.7% on HumanEvalFix, exceeding prior non-interactive approaches.
CodeThinker improves LLM code reasoning via consistency-based RL with stepwise training data, dynamic beam sampling, and consistency rewards, reaching SOTA on benchmarks with 4.3% gains on Qwen2.5-Coder-7B.
Solvita is an agentic evolution system using Planner, Solver, Oracle, and Hacker agents with trainable graph knowledge networks updated by reinforcement learning on pass/fail and vulnerability signals to achieve SOTA code generation performance.
DARE co-evolves difficulty estimation and policy in RL for LLMs to improve training efficiency, final performance, and inference speed by using tailored strategies for different difficulty levels.
Introduces a three-fold benchmark for LLM-generated code combining correctness testing on a complex project, quality verification, and developer surveys to assess production readiness.
Semantic distance on program execution behaviors improves uncertainty estimation for LLM code generation and outperforms prior sample-based methods across benchmarks and models.
A review of 114 studies creates taxonomies for code and data quality issues, formalizes 18 propagation mechanisms from training data defects to LLM-generated code defects, and synthesizes detection and mitigation techniques.
SpecValidator detects lexical vagueness, under-specification, and syntax-formatting defects in LLM code-generation prompts with F1 0.804, outperforming GPT-5-mini and Claude Sonnet 4, and shows that under-specification is the most damaging defect type while richer benchmarks are more resilient.
DryRUN lets LLMs create their own test inputs and run internal simulations for self-correcting code generation, matching the performance of test-dependent methods like CodeSIM on LiveCodeBench without public tests or external signals.
BAR trains independent domain experts via separate mid-training, SFT, and RL pipelines then composes them with a MoE router to match monolithic retraining performance at lower cost and without catastrophic forgetting.
Co-locating tests with implementation code yields substantially higher preservation and correctness in foundation-model-generated programs than separated test syntax.
Dual Reasoning with explicit safety audits improves the new SUDS metric by 1.32x to 3.42x over baselines on code generation benchmarks containing injected harmful keywords.
Large language models derive exact analytical GPU thread mappings for complex 2D/3D domains and fractals via in-context learning, outperforming symbolic regression and enabling up to thousands-fold speedups and energy reductions.
Ensemble Semantic Entropy improves correlation with code correctness over single-model methods and powers a cascading scaling system that cuts FLOPs by 64.9% while preserving performance on LiveCodeBench.
Mercury Coder diffusion LLMs achieve throughputs of 1109 and 737 tokens per second on H100 GPUs, up to 10x faster than frontier models with comparable quality.
LiveCodeBench collects 400 recent contest problems to create a contamination-free benchmark evaluating LLMs on code generation and related capabilities like self-repair and execution.
A 1.3B-parameter code model trained on 7B tokens of curated textbook and synthetic data achieves 50.6% on HumanEval, indicating data quality can enable strong performance at small scale.
CAMEL proposes a role-playing framework with inception prompting that enables autonomous multi-agent cooperation among LLMs and generates conversational data for studying their behaviors.
citing papers explorer
-
SWE-bench: Can Language Models Resolve Real-World GitHub Issues?
SWE-bench reveals that even top language models like Claude 2 resolve only 1.96% of 2,294 real-world GitHub issues, highlighting a gap in practical coding capabilities.
-
Mercury: Ultra-Fast Language Models Based on Diffusion
Mercury Coder diffusion LLMs achieve throughputs of 1109 and 737 tokens per second on H100 GPUs, up to 10x faster than frontier models with comparable quality.
-
Textbooks Are All You Need
A 1.3B-parameter code model trained on 7B tokens of curated textbook and synthetic data achieves 50.6% on HumanEval, indicating data quality can enable strong performance at small scale.
-
Learning to Foresee: Unveiling the Unlocking Efficiency of On-Policy Distillation
On-policy distillation gains efficiency from early foresight in module allocation and update directions, which the proposed EffOPD method exploits for 3x faster training with comparable performance.
-
NVIDIA Nemotron 3: Efficient and Open Intelligence
NVIDIA releases the Nemotron 3 model family with hybrid Mamba-Transformer architecture, LatentMoE, NVFP4 training, MTP layers, and multi-environment RL post-training for reasoning and agentic tasks.
-
Phi-4-Mini Technical Report: Compact yet Powerful Multimodal Language Models via Mixture-of-LoRAs
Phi-4-Mini achieves strong math and coding performance with only 3.8B parameters via high-quality synthetic data, while Phi-4-Multimodal uses Mixture-of-LoRAs to integrate modalities and top speech recognition leaderboards.
-
StarCoder: may the source be with you!
StarCoderBase matches or beats OpenAI's code-cushman-001 on multi-language code benchmarks; the Python-fine-tuned StarCoder reaches 40% pass@1 on HumanEval while retaining other-language performance.
-
Mellum2 Technical Report
Mellum 2 is a 12B MoE model with 2.5B active parameters, trained on 10.6T tokens with MoE, GQA, SWA, and MTP, then post-trained into Instruct and Thinking variants, claimed competitive with 4B-14B models at 2.5B compute.