Skip to main content

Vals CyberBench (CyberBench)

Can autonomous agents craft PoC inputs that trigger OSS-Fuzz vulnerabilities—and stop crashing after the fix?

Data verified

How BenchLM shows CyberBench

BenchLM mirrors the public Vals AI CyberBench leaderboard captured from https://www.vals.ai/benchmarks/cyber and updated by Vals on July 9, 2026. The snapshot preserves overall scores, uncertainty, latency, cost-per-test metadata, and task-level scores where Vals publishes them.

CyberBench is display only on BenchLM. Vals proprietary or Vals-hosted aggregate views are useful context, but BenchLM does not use them as weighted ranking inputs or as a replacement for benchmark-native source records.

14 Vals rows3 task viewsprivate datasetTasks: Overall, PoC, PatchDisplay only

CyberBench score on CyberBench — July 9, 2026

BenchLM mirrors the published cyberbench score view for CyberBench. GPT-5.6 Sol leads the public snapshot at 88.14% , followed by GPT-5.5 (80.51%) and GLM 5.2 (77.36%). BenchLM does not use these results to rank models overall.

14 modelsExternal benchmark mirrorsCurrentDisplay onlyUpdated July 9, 2026

The published CyberBench snapshot is tightly clustered at the top: GPT-5.6 Sol sits at 88.14%, while the third row is only 10.78 points behind. The broader top-10 spread is 35.99 points, so the benchmark still separates strong models even when the leaders cluster.

14 models have been evaluated on CyberBench. The benchmark falls in the External benchmark mirrors category. BenchLM tracks this category separately from its weighted global scoring system, so these results are best compared on the dedicated Korean benchmark views. CyberBench is currently displayed for reference but excluded from the scoring formula, so it does not directly affect overall rankings.

About CyberBench

Year

2026

Tasks

OSS-Fuzz PoC and patch-verification tasks

Format

Accuracy score

Difficulty

Autonomous cybersecurity exploit reproduction

BenchLM mirrors the public Vals AI CyberBench leaderboard as display-only external evidence. The captured snapshot preserves overall scores, task-level scores where Vals publishes them, uncertainty, latency, and cost-per-test metadata. It is excluded from BenchLM weighted rankings.

BenchLM freshness & provenance

Version

CyberBench 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.

CyberBench score table (14 models)

1
GPT-5.6 Solopenai/gpt-5.6-sol
88.14%
2
GPT-5.5openai/gpt-5.5
80.51%
3
GLM 5.2zai/glm-5.2
77.36%
4
MiniMax M3minimax/MiniMax-M3
76.27%
5
Kimi K2.6kimi/kimi-k2.6
74.35%
6
GPT-5.4openai/gpt-5.4-2026-03-05
72.88%
7
Gemini 3.5 Flashgoogle/gemini-3.5-flash
69.33%
8
DeepSeek V4 Prodeepseek/deepseek-v4-pro
68.64%
9
DeepSeek V4 Flashdeepseek/deepseek-v4-flash
66.72%
10
Claude Opus 4.7anthropic/claude-opus-4-7
52.14%
11
Claude Opus 4.8anthropic/claude-opus-4-8
50.85%
12
Grok 4.3grok/grok-4.3
47.46%
13
Qwen3.7 Plusalibaba/qwen3.7-plus
37.50%
14
Gemini 3.1 Pro Previewgoogle/gemini-3.1-pro-preview
36.44%

FAQ

What does CyberBench measure?

Can autonomous agents craft PoC inputs that trigger OSS-Fuzz vulnerabilities—and stop crashing after the fix?

Which model leads the published CyberBench snapshot?

GPT-5.6 Sol currently leads the published CyberBench snapshot with 88.14% cyberbench score. BenchLM shows this benchmark for display only and does not use it in overall rankings.

How many models are evaluated on CyberBench?

14 AI models are included in BenchLM's mirrored CyberBench snapshot, based on the public leaderboard captured on July 9, 2026.

Last updated: July 9, 2026 · mirrored from the public benchmark leaderboard

The AI models change fast. We track them for you.

For engineers, researchers, and the plain curious — a weekly brief on new models, ranking shifts, and pricing changes.

Free. No spam. Unsubscribe anytime.