mufeedvh

N-Day-Bench – Can LLMs find real vulnerabilities in real codebases?

N-Day-Bench tests whether frontier LLMs can find known security vulnerabilities in real repository code. Each month it pulls fresh cases from GitHub security advisories, checks out the repo at the last commit before the patch, and gives models a sandboxed bash shell to explore the codebase.

Static vulnerability discovery benchmarks become outdated quickly. Cases leak into training data, and scores start measuring memorization. The monthly refresh keeps the test set ahead of contamination — or at least makes the contamination window honest.

Each case runs three agents: a Curator reads the advisory and builds an answer key, a Finder (the model under test) gets 24 shell steps to explore the code and write a structured report, and a Judge scores the blinded submission. The Finder never sees the patch. It starts from sink hints and must trace the bug through actual code.

Only repos with 10k+ stars qualify. A diversity pass prevents any single repo from dominating the set. Ambiguous advisories (merge commits, multi-repo references, unresolvable refs) are dropped.

Currently evaluating GPT-5.4, Claude Opus 4.6, Gemini 3.1 Pro, GLM-5.1, and Kimi K2.5. All traces are public.

Methodology: https://ndaybench.winfunc.com/methodology

Live Leaderboard: https://ndaybench.winfunc.com/leaderboard

Live Traces: https://ndaybench.winfunc.com/traces

https://ndaybench.winfunc.com

N-Day-Bench

Strict GitHub advisory benchmarking with OpenRouter-backed finder models.

Apr 13 at 9:54pmWeb
Show threadRohinator2d ago
Very curious how Claude Mythos will perform here
Show threadmbbutler2d ago
It would be helpful to add in some cases that do not contain any vulnerabilities to assess false-positive rate as well.
Show threadmufeedvh2d ago

This is a good idea.

Will incorporate false-positive rates into the rubric from the next run onwards.

At winfunc, we spent a lot of research time taming these models to eradicate false-positive rates (it's high!) so this does feel important enough to be documented. Thanks!

Show threadcortesoft2d ago
Any code that is certain that it doesn't have any vulnerabilities is going to be pretty trivial to verify.
Show threadCynddl2d ago

> Each case runs three agents: a Curator reads the advisory and builds an answer key, a Finder (the model under test) gets 24 shell steps to explore the code and write a structured report, and a Judge scores the blinded submission. The Finder never sees the patch. It starts from sink hints and must trace the bug through actual code.

Curator, answer key, Finder, shell steps, structured report, sink hints… I understand nothing. Did you use an LLM to generate this HN submission?

It looks like a standard LLM-as-a-judge approach. Do you manually validate or verify some of the results? Done poorly, the results can be very noisy and meaningless.

Show threadpeyton2d ago

> Did you use an LLM to generate this HN submission?

Must have.

> The Finder will never see the patch.

I wasn’t worried that this eval would show the answer to the model before evaluating it. Seems requirements leaked into this post.

Show threadrohansood152d ago

I worked in AppSec in the past, made sense to me. Maybe you aren't the target audience?

You don't really need manual verification for these, the CVEs (vulnerabilities) are public and can be programmatically validated.

Show threadaos_architect2d ago
[dead]
Show threadvolume_tech2d ago
[dead]