# Vals CyberBench (CyberBench)

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

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

- Category: [Agentic](/agentic)
- Last updated: October 12, 2026

Evaluation results published by [Vals AI](https://www.vals.ai/benchmarks/cyber). Published leaderboard aggregates and evaluation settings only. Dataset contents are not included.

## About CyberBench

- Year: 2026
- Tasks: OSS-Fuzz PoC and patch-verification tasks
- Format: Accuracy score
- Difficulty: Autonomous cybersecurity exploit reproduction
- Paper: [CyberBench](https://www.vals.ai/benchmarks/cyber)

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.

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

## Leaderboard (47 models)

| Rank | Model | Configuration | Creator | Score |
|------|-------|---------------|---------|-------|
| 1 | [Muse Wonder](/models/muse-wonder) | max reasoning | Meta | 80.36% |
| 2 | [GPT-6 Sol](/models/gpt-6-sol) | max reasoning | OpenAI | 77.98% |
| 3 | [Gemini 4 Argon](/models/gemini-4-argon) | high reasoning | Google | 77.86% |
| 4 | [GPT-5.6 Sol](/models/gpt-5-6-sol) | max reasoning | OpenAI | 76.31% |
| 5 | [GPT-6 Luna](/models/gpt-6-luna) | max reasoning | OpenAI | 76.25% |
| 6 | [Claude Haiku 5.5](/models/claude-haiku-5-5) | — | Anthropic | 75.48% |
| 7 | [MiMo-V2.6-Flash](/models/mimo-v2-6-flash) | — | Xiaomi | 75.36% |
| 8 | [Kimi K3](/models/kimi-k3) | max reasoning | Moonshot AI | 75.24% |
| 9 | [Claude Opus 5.5 Range](https://www.vals.ai/models/anthropic_claude-opus-5-5-range) | — | Anthropic | 74.58% |
| 10 | [DeepSeek V4.1 Flash 1017](/models/deepseek-v4-1-flash) | high reasoning | DeepSeek | 73.69% |
| 11 | [GPT-5.6 Luna](/models/gpt-5-6-luna) | max reasoning | OpenAI | 73.63% |
| 12 | [Muse Wonder Full](/models/muse-wonder-full) | max reasoning | Meta | 73.63% |
| 13 | [GPT Daybreak Blue](https://www.vals.ai/models/openai_gpt-daybreak-blue) | max reasoning | OpenAI | 72.98% |
| 14 | [DeepSeek V4 Pro 0813](/models/deepseek-v4-pro-0813) | max reasoning | DeepSeek | 72.86% |
| 15 | [MiMo-V2.6-Pro](/models/mimo-v2-6-pro) | — | Xiaomi | 72.86% |
| 16 | [Muse Spark 1.3 Max](https://www.vals.ai/models/meta_muse_spark_1_3_max) | max reasoning | Meta | 72.74% |
| 17 | [DeepSeek V4 Flash 0731](/models/deepseek-v4-flash-0731) | high reasoning | DeepSeek | 72.08% |
| 18 | [GPT-5.6 Terra](/models/gpt-5-6-terra) | max reasoning | OpenAI | 72.08% |
| 19 | [GLM-5.3](/models/glm-5-3) | max reasoning | Z.AI | 72.08% |
| 20 | [Grok 4.5](/models/grok-4-5) | high reasoning | xAI | 71.01% |
| 21 | [Claude Fable 5.1](/models/claude-fable-5-1) | — | Anthropic | 70.42% |
| 22 | [Claude Sonnet 5 Range](https://www.vals.ai/models/anthropic_claude-sonnet-5-range) | — | Anthropic | 70.30% |
| 23 | [Muse Spark 1.2](/models/muse-spark-1-2) | xhigh reasoning | Meta | 69.52% |
| 24 | [Ember-1](/models/ember-1) | — | Fireworks | 69.52% |
| 25 | [Grok 4.7](/models/grok-4-7) | xhigh reasoning | xAI | 69.46% |
| 26 | [Muse Spark 1.3](/models/muse-spark-1-3) | xhigh reasoning | Meta | 69.41% |
| 27 | [GLM-5.3-Flash](/models/glm-5-3-flash) | max reasoning | Z.AI | 67.86% |
| 28 | [Qwen3.8-27B](/models/qwen3-8-27b) | xhigh reasoning | Alibaba | 67.80% |
| 29 | [Grok 4.6](/models/grok-4-6) | high reasoning | xAI | 67.68% |
| 30 | [Inkling](/models/inkling) | 0.99 reasoning | Thinking Machines Lab | 66.79% |
| 31 | [Claude Opus 5](/models/claude-opus-5) | — | Anthropic | 65.36% |
| 32 | [Hy4 preview](/models/hy4-preview) | — | Tencent | 65.36% |
| 33 | [MiniMax M3](/models/minimax-m3) | — | MiniMax | 65.00% |
| 34 | [Inkling-Small](/models/inkling-small) | 0.99 reasoning | Thinking Machines Lab | 64.11% |
| 35 | [Claude Sonnet 5](/models/claude-sonnet-5) | — | Anthropic | 61.91% |
| 36 | [Claude Sonnet 5.5](/models/claude-sonnet-5-5) | — | Anthropic | 59.58% |
| 37 | [Gemini 3.5 Flash-Lite](/models/gemini-3-5-flash-lite) | high reasoning | Google | 59.11% |
| 38 | [Ling 3.0 Flash Af Rc3](https://www.vals.ai/models/ant_ling-3.0-flash-af-rc3) | — | Ant | 58.33% |
| 39 | [Claude Opus 5.5](/models/claude-opus-5-5) | — | Anthropic | 55.36% |
| 40 | [Mercury 2.5](/models/mercury-2-5) | high reasoning | Inception | 45.48% |
| 41 | [Gemini 3.6 Flash](/models/gemini-3-6-flash) | high reasoning | Google | 45.36% |
| 42 | [Gemini 3.7 Flash](/models/gemini-3-7-flash) | high reasoning | Google | 43.75% |
| 43 | [Gemini 3.8 Flash](/models/gemini-3-8-flash) | high reasoning | Google | 43.75% |
| 44 | [GPT-6 Astra](/models/gpt-6-astra) | max reasoning | OpenAI | 41.07% |
| 45 | [GPT-6.1 Sol](/models/gpt-6-1-sol) | max reasoning | OpenAI | 39.29% |
| 46 | [Nemotron Lightning 3p5 30b A3b](https://www.vals.ai/models/fireworks_nemotron-lightning-3p5-30b-a3b) | — | Fireworks AI | 29.46% |
| 47 | [Qwen3.8 Max](/models/qwen3-8-max) | — | Alibaba | 28.57% |

## 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?

Muse Wonder currently leads the published CyberBench snapshot with a score of 80.36%.

### How many models are evaluated on CyberBench?

The October 12, 2026 contains 47 AI models.

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

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