# Cybench

> A cybersecurity benchmark of professional Capture the Flag tasks for measuring autonomous cyber agent capability and risk.

Canonical page: https://benchlm.ai/benchmarks/cybench

- Category: [Agentic](/agentic)
- Last updated: October 17, 2026

## About Cybench

- Year: 2025
- Tasks: 40 professional CTF tasks
- Format: Cybersecurity agent task completion
- Difficulty: Professional cybersecurity
- Paper: [Cybench: A Framework for Evaluating Cybersecurity Capabilities and Risks of Language Models](https://arxiv.org/abs/2408.08926)

Cybench packages recent professional CTF tasks with execution environments and subtasks. BenchLM tracks provider-reported Cybench rows as display-only security-agent capability signals.

Cybench is currently displayed on BenchLM for reference, but it is excluded from the weighted scoring formula.

## Leaderboard (2 models)

| Rank | Model | Creator | Score |
|------|-------|---------|-------|
| 1 | [Mistral Large 4](/models/mistral-large-4) | Mistral | 93.0% |
| 2 | [Muse Spark 1.1](/models/muse-spark-1-1) | Meta | 92.9% |

## FAQ

### What does Cybench measure?

A cybersecurity benchmark of professional Capture the Flag tasks for measuring autonomous cyber agent capability and risk.

### Which model scores highest on Cybench?

Mistral Large 4 by Mistral currently leads with a score of 93.0% on Cybench.

### How many models are evaluated on Cybench?

2 AI models have published results on Cybench in the BenchLM catalog.

### Does Cybench affect BenchLM's overall score?

Not directly. Cybench is still displayed on BenchLM for reference, but it is excluded from the weighted scoring formula.

## Compare Top Models on Cybench

- [Mistral Large 4 vs Muse Spark 1.1](/compare/mistral-large-4-vs-muse-spark-1-1)