Atlas
Benchmarks
331 sources · 1,258 models × 2,980 benchmarks · 155,643 scoresWiz Cyber Model Arena v3.9 (Cloud Security)
The Cloud Security category of Wiz's Cyber Model Arena v3.9: 23 capture-the-flag challenges taken from Wiz's CTF challenges, in which an agent exploits misconfigurations to capture a flag in AWS, Azure, GCP or Kubernetes-style setups. Pass@1 averages three attempts per challenge, for a model in a general-purpose harness.
Models
24
Top score
87.0
Leader
Gemini 4 Argon
Updated
—
Resolution
—
Saturation
—
No signal or difficulty estimate. Held out of the fit by design.
Organizations
| # | Model | Setting | Score | Source |
|---|---|---|---|---|
| 1 | Gemini 4 Argon | high · Antigravity CLI | 87.0 | |
| 2 | GPT-6 Astra | high · Claude Code | 72.5 | |
| 3 | GPT-6.1 Sol | high · Codex | 68.1 | |
| 4 | Gemini 3.8 Flash Cyber | high · Wiz Cyber Model Arena ADK React; simple ReAct loop built with ADK; single bash tool; no cyber-specific tools; general system prompt | 66.7 | |
| 5 | Gemini 3.7 Flash | high · Antigravity CLI | 65.2 | |
| 6 | GPT-5.6 Sol | high · Codex | 65.2 | |
| 7 | GPT-6 Sol | high · Codex | 65.2 | |
| 8 | Opus 4.8 | high · Claude Code | 63.8 | |
| 9 | Claude Opus 5 | high · Claude Code | 63.8 | |
| 10 | DeepSeek V4.1 Flash 1017 | high · Claude Code | 62.3 | |
| 11 | GPT-5.6 Terra | high · Codex | 62.3 | |
| 12 | Gemini 3.6 Flash | high · Antigravity CLI | 58.0 | |
| 13 | GPT-6 Luna | high · Codex | 58.0 | |
| 14 | Sonnet 5 | high · Claude Code | 56.5 | |
| 15 | DeepSeek V4.1 Flash | high · Claude Code | 56.5 | |
| 16 | DeepSeek V4 Pro 0813 | high · Wiz Cyber Model Arena ADK React; simple ReAct loop built with ADK; single bash tool; no cyber-specific tools; general system prompt | 53.6 | |
| 17 | Claude Haiku 5.5 | high · Claude Code | 50.7 | |
| 18 | Sonnet 4.5 | high · Claude Code | 50.7 | |
| 19 | GLM-5.3 | high · Claude Code | 50.7 | |
| 20 | Gemini 3.5 Flash | high · Claude Code | 49.3 | |
| 21 | Kimi K3 | high · Wiz Cyber Model Arena ADK React; simple ReAct loop built with ADK; single bash tool; no cyber-specific tools; general system prompt | 49.3 | |
| 22 | GLM-5.3 Flash | high · Claude Code | 42.0 | |
| 23 | Grok 4.6 | high · Wiz Cyber Model Arena ADK React; simple ReAct loop built with ADK; single bash tool; no cyber-specific tools; general system prompt | 39.1 | |
| 24 | MiMo-V2.6-Pro | high · Claude Code | 36.2 |
