gemini-4IMAGE

Gemini 4 Argon scores 87% on ExploitBench and 68% on CWE-Bench

A leak says Argon reaches 87% on exploit generation, behind Opus 5.5's 91%, and ties GPT-6 Astra for first at 68% on vulnerability discovery, which the post uses to explain Google's release limits.

xiax｜AI & Agents

@xiax0603Archived 20 days agoOriginally posted 2026.10.14

Gemini 4 Argon scores 87% on ExploitBench and 68% on CWE-Bench (image 1 of 2)

Gemini 4 Argon scores 87% on ExploitBench and 68% on CWE-Bench (image 2 of 2)

xiax｜AI & Agents @xiax0603Originally posted 2026.10.14

Open original on X

🚨Gemini 4 Argon：顶级攻防Agent能力，漏洞挖掘基准成绩单！两份前沿安全Agent基准，ExploitBench与CWE‑Bench，能看到Argon的网络化能力有多突出。 🔹ExploitBench（漏洞利用） Argon拿到87%，仅次于Claude Opus5.5的91%，大幅甩开Meta Muse Wonder。 🔹CWE‑Bench v1（漏洞发现、代码缺陷修复） Argon 68%，和GPT‑6 Astra并列榜单第一，高于Opus5.5的67%。结合Apollo报告前面的结论：高GCI综合能力 + 超标的隐秘规避能力 + 强烈评测感知 + 顶尖攻防Agent实力。这就是为什么谷歌对Argon做严格的发布限制。

This post quotes another post: x.com/xiax0603/status/2110198122236510620

x.com/xiax0603/status/2110202586414579798Archived 2026.10.15 11:39