Google推出了Gemini 3.8 Flash和Gemini 3.8 Flash Cyber两款新型AI模型1。其中Gemini 3.8 Flash是一款改进的推理与编码模型，在软件工程、自主代理任务和多步推理方面相比3.7版本实现了显著提升，在HLE-Verified基准测试中达到54.9%1。该模型的定价与3.7 Flash保持一致，输入成本为每百万tokens 0.75美元，输出成本为每百万tokens 3.75美元1。
Gemini 3.8 Flash Cyber是一款专为网络安全优化的模型，仅通过Fairwind项目提供给受信任的防御者1。该模型在漏洞检测与自动修补方面具有业界领先的能力1。在CWE-Bench上的pass@1达到47.2%，相比领先的前沿模型47.8%具有成本优势1。Google云漏洞研究团队使用3.8 Flash Cyber在2小时内发现了关键基础漏洞，而此类发现通常需要数月研究1。Chrome安全团队发现该模型生成的正确补丁数是最好商业模型的2.6倍1。Wiz在内部渗透测试基准上发现其召回率提升7.5-9.7%，同时成本降低2.3-5.2倍1。在Gray Swan测试中，该模型的提示注入鲁棒性得到了显著提升1。
Google has introduced two new artificial intelligence models: Gemini 3.8 Flash, an enhanced general-purpose model for reasoning and coding, and Gemini 3.8 Flash Cyber, a specialized variant optimized for cybersecurity applications.1 Gemini 3.8 Flash represents a significant advancement over its predecessor, with marked improvements in software engineering, autonomous agent tasks, and multi-step reasoning capabilities.1 The pricing for Gemini 3.8 Flash remains unchanged from the previous generation, at $0.75 per million input tokens and $3.75 per million output tokens.1
The cybersecurity-focused variant, Gemini 3.8 Flash Cyber, delivers industry-leading vulnerability detection and automated patching capabilities.1 According to benchmarks, Chrome's security team found that patches generated by Gemini 3.8 Flash Cyber were 2.6 times more effective than those from the best competing commercial models.1 On the CWE-Bench assessment, the model achieved a pass rate of 47.2%, delivering cost advantages compared to leading frontier models with comparable performance at 47.8%.1 Independent testing by Wiz demonstrated that Gemini 3.8 Flash Cyber achieved 7.5 to 9.7 percent higher recall rates on internal penetration testing benchmarks while reducing costs by 2.3 to 5.2 times.1 Google's vulnerability research team used Gemini 3.8 Flash Cyber to identify critical infrastructure vulnerabilities in two hours, a discovery process that typically requires months of research.1 The model also demonstrates significant robustness improvements against prompt injection attacks in Gray Swan testing.1 Gemini 3.8 Flash Cyber is currently available exclusively through the Fairwind initiative to trusted defenders.1
评论
还没有评论，欢迎留下第一条。