CyberBench leaderboard
As of 6 Nov 2026, the top score on CyberBench is 80.4% by Muse Wonder (Meta), ahead of GPT-6 Sol at 78.0%, across 48 model configurations with a published result (Vals AI, result dated 5 Nov 2026). Given a codebase with a vulnerability, the agent must find it and patch it so the exploit no longer works and the tests still pass. CyberBench is shown for reference only and does not count toward the BenchLeader Index.
Key facts (as of 6 Nov 2026)
- Muse Wonder holds the top CyberBench score at 80.4% (Vals AI, 5 Nov 2026).
- GPT-6 Sol is second on CyberBench at 78.0% (Vals AI, 5 Nov 2026).
- CyberBench: Finding and patching security bugs in real code. Run by Vals AI (Vals AI).
- 48 model configurations have a published CyberBench result on BenchLeader (BenchLeader, 6 Nov 2026).
- CyberBench does not feed the BenchLeader Index; it is shown for reference (BenchLeader methodology).
Finding and patching security bugs in real code. Run by Vals AI.
- Published by
- Vals AI
- Category
- Agents & tools
- Index weight
- Reference only
- Models
- 48
- Data as of
- 6 Nov 2026
Vals AI (vals.ai).
What the test looks like
Given a codebase with a vulnerability, the agent must find it and patch it so the exploit no longer works and the tests still pass.
How it is scored
Percent of vulnerabilities fixed, run by Vals AI.
What to keep in mind
Complements Cybench's capture-the-flag framing with defensive work.
- 1Muse Wonder (max)80.4%
- 2GPT-6 Sol (max)78.0%
- 3Gemini 4 Argon (high)77.9%
- 4GPT-5.6 Sol (max)76.3%
- 5GPT-6 Luna (max)76.3%
- 6Claude Haiku 5.575.5%
- 7MiMo-V2.6-Flash75.4%
- 8Kimi K3 (max)75.2%
- 9Claude Opus 5.5 Range74.6%
- 10DeepSeek V4.1 Flash (high)73.7%
- 11Muse Wonder Full (max)73.6%
- 12GPT-5.6 Luna (max)73.6%
- 13Daybreak Blue (max)73.0%
- 14MiMo-V2.6-Pro72.9%
- 15DeepSeek V4 Pro (max)72.9%
48 of 48
| # | |||||
|---|---|---|---|---|---|
| 1 |  Muse Wondermax | 80.4% | 62.3 | $3.66 | 2026-11-05 |
| 2 |  GPT-6 Solmax | 78.0% | 63.3 | $1.34 | 2026-11-05 |
| 3 |  Gemini 4 Argonhigh | 77.9% | 66.0 | $5.11 | 2026-11-05 |
| 4 | GPT-5.6 Solmax | 76.3% | 65.4 | $3.05 | 2026-11-05 |
| 5 | GPT-6 Lunamax | 76.3% | 56.7 | $0.134 | 2026-11-05 |
| 6 |  Claude Haiku 5.5 | 75.5% | – | $1.25 | 2026-11-05 |
| 7 |  MiMo-V2.6-Flashopen ↗ | 75.4% | 55.0 | $0.046 | 2026-11-05 |
| 8 |  Kimi K3maxopen ↗ | 75.2% | 62.1 | $2.13 | 2026-11-05 |
| 9 | Claude Opus 5.5 Range | 74.6% | – | $4.17 | 2026-11-05 |
| 10 |  DeepSeek V4.1 Flashhighopen ↗ | 73.7% | 55.7 | $0.105 | 2026-11-05 |
| 11 | Muse Wonder Fullmax | 73.6% | 62.9 | $3.63 | 2026-11-05 |
| 12 | GPT-5.6 Lunamax | 73.6% | 56.3 | $0.404 | 2026-11-05 |
| 13 | Daybreak Bluemax | 73.0% | – | $3.97 | 2026-11-05 |
| 14 | MiMo-V2.6-Proopen ↗ | 72.9% | 57.3 | $0.094 | 2026-11-05 |
| 15 | DeepSeek V4 Promaxopen ↗ | 72.9% | 57.0 | $0.593 | 2026-11-05 |
| 16 | Muse Spark 1.3max | 72.7% | 62.0 | $3.34 | 2026-11-05 |
| 17 | NewInkling 2 Small | 72.7% | 57.4 | $1.27 | 2026-11-05 |
| 18 | GPT-5.6 Terramax | 72.1% | 60.7 | $3.31 | 2026-11-05 |
| 19 |  GLM 5.3maxopen ↗ | 72.1% | 59.0 | $2.69 | 2026-11-05 |
| 20 | DeepSeek V4 Flashhighopen ↗ | 72.1% | 57.3 | $0.193 | 2026-11-05 |
| 21 |  Grok 4.5high | 71.0% | 56.9 | $2.95 | 2026-11-05 |
| 22 | Claude Fable 5.1 | 70.4% | 66.4 | $4.01 | 2026-11-05 |
| 23 | Claude Sonnet 5 Range | 70.3% | – | $4.84 | 2026-11-05 |
| 24 | Muse Spark 1.2xhigh | 69.5% | 61.7 | $2.15 | 2026-11-05 |
| 25 | Ember-1 | 69.5% | – | $2.43 | 2026-11-05 |
| 26 | Grok 4.7xhigh | 69.5% | 58.3 | $8.54 | 2026-11-05 |
| 27 | Muse Spark 1.3xhigh | 69.4% | 58.5 | $4.26 | 2026-11-05 |
| 28 | GLM 5.3 Flashmaxopen ↗ | 67.9% | 54.2 | $0.284 | 2026-11-05 |
| 29 |  Qwen3.8 27Bxhighopen ↗ | 67.8% | 52.5 | $0.990 | 2026-11-05 |
| 30 | Grok 4.6high | 67.7% | 60.5 | $4.07 | 2026-11-05 |
| 31 | Inklingopen ↗ | 66.8% | 47.5 | $1.46 | 2026-11-05 |
| 32 | Claude Opus 5 | 65.4% | 65.2 | $3.53 | 2026-11-05 |
| 33 |  Hy4open ↗ | 65.4% | 55.7 | $0.458 | 2026-11-05 |
| 34 |  MiniMax M3open ↗ | 65.0% | 56.4 | $3.41 | 2026-11-05 |
| 35 | Inkling Smallopen ↗ | 64.1% | 52.1 | $0.252 | 2026-11-05 |
| 36 | Claude Sonnet 5 | 61.9% | 55.9 | $1.89 | 2026-11-05 |
| 37 | Claude Sonnet 5.5 | 59.6% | – | $4.42 | 2026-11-05 |
| 38 | Gemini 3.5 Flash Litehigh | 59.1% | 47.9 | $0.425 | 2026-11-05 |
| 39 |  Ling 3.0 Flash Af Rc3 | 58.3% | – | $0.104 | 2026-11-05 |
| 40 | Claude Opus 5.5 | 55.4% | 63.4 | $3.48 | 2026-11-05 |
| 41 |  Mercury 2.5high | 45.5% | – | $0.125 | 2026-11-05 |
| 42 | Gemini 3.6 Flashhigh | 45.4% | 61.2 | $1.31 | 2026-11-05 |
| 43 | Gemini 3.7 Flashhigh | 43.8% | 62.2 | $2.15 | 2026-11-05 |
| 44 | Gemini 3.8 Flashhigh | 43.8% | 62.0 | $1.96 | 2026-11-05 |
| 45 | GPT-6 Astramax | 41.1% | 67.8 | $1.84 | 2026-11-05 |
| 46 | GPT-6.1 Solmax | 39.3% | 67.3 | $0.285 | 2026-11-05 |
| 47 |  Nemotron Lightning 3.5 30B A3B | 29.5% | – | $0.093 | 2026-11-05 |
| 48 | Qwen3.8 Maxmax | 28.6% | 58.8 | $0.785 | 2026-11-05 |
More agents & tools benchmarks · All LLM benchmarks
Related benchmarks
Top head-to-heads on CyberBench
Cite as: BenchLeader, “CyberBench leaderboard”, data as of 6 Nov 2026, https://www.benchleader.com/benchmarks/vals_cyber.
CyberBench: questions
- What does CyberBench measure?
- Given a codebase with a vulnerability, the agent must find it and patch it so the exploit no longer works and the tests still pass. Scores are reported in percent of tasks solved; higher is better.
- What is the top score on CyberBench?
- As of 6 Nov 2026, the highest score on CyberBench is 80.4% by Muse Wonder (result dated 5 Nov 2026), ahead of GPT-6 Sol at 78.0%. Scores come from Vals AI and are re-read daily.
- Which AI model leads CyberBench?
- Muse Wonder leads CyberBench with 80.4% as of 6 Nov 2026 (result dated 5 Nov 2026), ahead of GPT-6 Sol at 78.0%.
- How many models have CyberBench results?
- 48 model configurations have a CyberBench result on BenchLeader, all taken from Vals AI.
- Who runs CyberBench and how often is it updated?
- CyberBench is published by Vals AI. BenchLeader re-reads the published results once a day and records the date each result was published.
- Does CyberBench count toward the BenchLeader Index?
- No. CyberBench is shown for reference but left out of the composite index.