LetTokenBurnLintEval: Benchmarking LLMs on Linter-Guided File Refactoring
LintEval: Benchmarking LLMs on Linter-Guided File Refactoring π π π Accepted at EMNLP 2026 Findings.
$ npm i -g lettokenburn-cli$ ltb bench info linteval
External harness. No Harbor adapter has been generated for this benchmark yet. `ltb bench info` prints the known links.
Runs happen on your machine with your own subscription; only the score summary is pushed. CLI docs
| Harness | Source | Confidence | Link | |||
|---|---|---|---|---|---|---|
| GPT-5.4-mini | 90.4 | N/A | Independent | medium | Oct 14, 2026 | arxiv.org |
| GPT-5.2-chat | 87.8 | N/A | Independent | medium | Oct 14, 2026 | arxiv.org |
| DeepSeek V3.2 | 73.0 | N/A | Independent | medium | Oct 14, 2026 | arxiv.org |
Every number links to where it was published. Low-confidence rows are muted and need per-cell verification.
LintEval: Benchmarking LLMs on Linter-Guided File Refactoring
π π π Accepted at EMNLP 2026 Findings. Paper: arXiv:2610.20027 Β· PDF A multi-language benchmark for evaluating whether LLMs can refactor code based on linter feedback while preserving functionality. Given a program plus the alerts reported by a static-analysis linter, a model must produce a revised program that (1) removes or reduces the reported alerts and (2) still passes all unit tests. Unlike code-generationβ¦