RustMizan - Line Localization: leaderboard
RustMizan (2026) evaluates LLM agents with shell access on 173 compilable crate-, file- and function-level Rust variants built from 42 RustSec memory-safety CVEs. This board is the micro-averaged F1 of the vulnerable source lines the agent names, in percent; higher is better.
Metric: Micro-averaged F1 (%) of the predicted vulnerable (file, line) pairs over the 173 compilable crate-, file- and function-level Rust variants of 42 RustSec memory-safety CVEs, unmutated, each analysed by a ReAct agent with shell access in an Inspect sandbox that returns a structured JSON verdict; higher is better. Source: arxiv.org. Saturation forecast: Around February 2028. 4 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | Claude Sonnet 4.6 | 23.2 | #161 |
| 2 | Gemini 3.1 Pro (Preview) | 23.1 | #94 |
| 3 | Qwen 3.6 Plus | 20.2 | #222 |
| 4 | GPT-5.4 | 17.5 | #123 |
No result here: #1 Claude Fable 5.2, #2 Gemini 4 Carbon, #5 Claude Opus 5.5.
Interactive version: theaggregate.ai/benchmark?slug=rustmizan-line-localization · How It Works · Data refreshed daily, snapshot 2026-11-13.