Skip to main content
Radar

Five or fewer confirmed AI changes, with original sources, on mornings when something changed.A free source-linked morning brief.

See the free Radar Brief

CWE-Bench

An external benchmark for evaluating whether coding agents can produce correct patches for real-world software vulnerabilities.

Data verified 27 confirmed releases in the last 30 daysSee the free Radar Brief

Benchmark score on CWE-Bench — September 2, 2026

We mirror the published score view for CWE-Bench. Gemini 3.8 Flash Cyber leads the public snapshot at 47.2%. We do not use these results to rank models overall.

1 modelAgenticCurrentDisplay onlyUpdated September 2, 2026

Benchmark score table (1 model)

Score
1
Gemini 3.8 Flash CyberGoogle · Closed
47.2%

About CWE-Bench

Year

2026

Tasks

Real-world vulnerability patching tasks

Format

Pass@1

Difficulty

Automated vulnerability remediation

CWE-Bench evaluates automated vulnerability patching. BenchLM stores exact provider-reported pass@1 results as display-only cybersecurity evidence until a stable benchmark-owned result artifact is available for the exact model setup.

BenchLM freshness & provenance

Version

CWE-Bench 2026

Refresh cadence

Quarterly

Staleness state

Current

Question availability

Public benchmark set

CurrentDisplay only

BenchLM uses freshness metadata to decide whether a benchmark should still be treated as a strong differentiator, a benchmark to watch, or a display-only reference. For the full scoring policy, see the BenchLM methodology page.

FAQ

What does CWE-Bench measure?

An external benchmark for evaluating whether coding agents can produce correct patches for real-world software vulnerabilities.

Which model scores highest on CWE-Bench?

Gemini 3.8 Flash Cyber by Google currently leads with a score of 47.2% on CWE-Bench.

How many models are evaluated on CWE-Bench?

1 AI models have been evaluated on CWE-Bench on BenchLM.

Last updated: September 2, 2026 · BenchLM version CWE-Bench 2026

Know when it’s worth switching models

The model to choose, the cheaper alternative, and the release we would wait on.

Read a sample issue

Join 2,000+ readers.

One email each week. Unsubscribe anytime.