Question 1

What is ExploitBench?

Accepted Answer

: A 'capability ladder' for security research, built by CMU researchers: the AI is given known bugs in Chrome's V8 engine and scored on how far it gets toward a working exploit inside a research sandbox — from understanding the patch to triggering a crash. Higher is better.
: GPT-6 Astra by OpenAI holds the best ExploitBench result among tracked models, at 100% (released Sep 3 2026). Higher scores are better on this benchmark.
: 1. GPT-6 Astra (OpenAI) — 100%; 2. GLM-5.3 (Z.ai) — 54.4%.
: 2 tracked models have a published ExploitBench score. Scores are the figures reported by each lab at that model's release, so this page is a record of results over time rather than a re-run leaderboard.
: GLM-5.3 by Z.ai is the highest-ranked model with downloadable weights on ExploitBench, scoring 54.4% at rank 2 overall.

Question 2

Which AI model scores highest on ExploitBench?

Accepted Answer

GPT-6 Astra by OpenAI holds the best ExploitBench result among tracked models, at 100% (released Sep 3 2026). Higher scores are better on this benchmark.

Question 3

What are the top 2 models on ExploitBench?

Accepted Answer

1. GPT-6 Astra (OpenAI) — 100%; 2. GLM-5.3 (Z.ai) — 54.4%.

Question 4

How many models have a published ExploitBench score?

Accepted Answer

2 tracked models have a published ExploitBench score. Scores are the figures reported by each lab at that model's release, so this page is a record of results over time rather than a re-run leaderboard.

Question 5

What is the best open model on ExploitBench?

Accepted Answer

GLM-5.3 by Z.ai is the highest-ranked model with downloadable weights on ExploitBench, scoring 54.4% at rank 2 overall.

ExploitBench

Rankings

ExploitBench — frequently asked questions