精选理由
微软的新安全模型MAI-Cyber-1-Flash，比Anthropic Mythos高出12个点，成本还减半，搞网络安全的可以盯着。
微软发布了其首个网络安全模型MAI-Cyber-1-Flash，在CyberGym漏洞基准测试中取得96%的分数。该分数比Anthropic的Mythos高12个百分点，且推理成本仅为后者的一半。微软AI CEO Mustafa Suleyman表示，token成本已成为防御方的真正约束。
原文 · The Rundown AI
LAUNCHED: Microsoft has released MAI-Cyber-1-Flash, its first cybersecurity model. MAI scores 96% o...
LAUNCHED: Microsoft has released MAI-Cyber-1-Flash, its first cybersecurity model. MAI scores 96% on the CyberGym vulnerability benchmark inside the company's MDASH agent harness, 12 points above Anthropic's Mythos at half the cost. Microsoft AI CEO Mustafa Suleyman: "Token cost is now the real constraint for defenders". 💬 1 🔄 0 ❤️ 4 👀 1643 📊 2 ⚡