Question 1

What is CVE-Bench?

Accepted Answer

: Can the AI exploit real, publicly known security holes in web applications, in a sandbox built to resemble a live service? A measure of raw capability: it is run without the safeguards a deployed model carries. Higher is better.
: Grok 4.6 by SpaceXAI holds the best CVE-Bench result among tracked models, at 39.8% (released Aug 12 2026). Higher scores are better on this benchmark.
: 1. Grok 4.6 (SpaceXAI) — 39.8%; 2. Grok 4.7 (SpaceXAI) — 37.7%; 3. Grok 4.5 (SpaceXAI) — 35.2%.
: 3 tracked models have a published CVE-Bench score. Scores are collected from published lab reports and benchmark sources, with recorded sources credited alongside the results. We do not re-run the evaluations.

Question 2

Which AI model scores highest on CVE-Bench?

Accepted Answer

Grok 4.6 by SpaceXAI holds the best CVE-Bench result among tracked models, at 39.8% (released Aug 12 2026). Higher scores are better on this benchmark.

Question 3

What are the top 3 models on CVE-Bench?

Accepted Answer

1. Grok 4.6 (SpaceXAI) — 39.8%; 2. Grok 4.7 (SpaceXAI) — 37.7%; 3. Grok 4.5 (SpaceXAI) — 35.2%.

Question 4

How many models have a published CVE-Bench score?

Accepted Answer

3 tracked models have a published CVE-Bench score. Scores are collected from published lab reports and benchmark sources, with recorded sources credited alongside the results. We do not re-run the evaluations.

CVE-Bench

Rankings

CVE-Bench — frequently asked questions