Z.ai:GLM-5.3在CyberGym基准测试中达84.5%
Z.ai GLM-5.3通过agentic微调在CyberGym漏洞基准测试中达到84.5%,超越GLM-5.2及专有模型,并触发安全审查。
原文链接详细分析
Z.ai将GLM-5.3在CyberGym vulnerability benchmark LLM performance测试中推至84.5%,大幅超越前代GLM-5.2并领先顶级专有模型。工程师仅通过GLM model agentic capabilities fine-tuning safety优化实现突破,未改动基础权重。该模型在发现和利用漏洞方面能力激增,促使Z.ai暂缓开放权重发布以进行安全测试。
DeepLearning.AI
@DeepLearningAIWe are an education technology company with the mission to grow and connect the global AI community.