{"id":91659,"topic":"ai","source":"finance.sina.cn","title":"阿里巴巴“小强模型”登顶CyberGym模型榜 - finance.sina.cn","url":"https://finance.sina.cn/stock/jdts/2026-09-28/detail-initkarm3874841.d.html?oid=800&vt=4&cid=76993&node_id=76993","url_hash":"8d38ae73ea0cbe2b7afd47663f324d4b8289f503","author":"","summary":"<a href=\"https://news.google.com/rss/articles/CBMisgFBVV95cUxQVFJ1S2pIWWNadkVEcHYweUQzdTI4d2ZqZjFoUmxOMXNXd29XWnhmNHV1T2NSQVpKeTdnYjRaeEdHMUNLMDBlcjZMZjZpd3RBSVVMMEdqZ3F1VEJ6TWM1bFZ1clo3OWk3djZONmZ2S2tBSWYxZFJ4dHY1Q1BNdENWZWJ5UF85czVyVnlNWWpfRWxjakxXY2ZXQTFZWHBfaW43NTVscFZwYlduSFduanVQU3Jn?oc=5\" target=\"_blank\">阿里巴巴“小强模型”登顶CyberGym模型榜</a>&nbsp;&nbsp;<font color=\"#6f6f6f\">finance.sina.cn</font>","content":"阿里巴巴“小强模型”登顶CyberGym模型榜\n（来源：环球网资讯）\n来源：环球网\n【环球网科技综合报道】日前，阿里巴巴伏渊息壤大模型（XekRung-27B）以88.9%的成功率登上CyberGym榜首，成为该国际权威AI安全能力榜单历史上第一个来自中国的冠军模型。伏渊息壤在获得最高成功率的同时，还实现了“既小又强”的突破，27B的参数量仅为其他上榜模型的1/27~1/370，这意味着该“小强模型”极大降低了智能排查安全漏洞的算力成本。\nCyberGym是目前国际上最权威的AI安全能力榜单之一，由美国加州大学伯克利分校专业研究团队运营，其背后是一套AI安全能力基准测试框架，涵盖188个软件项目的1507个真实安全漏洞。Cybergym榜单分为模型榜单和智能体榜单两部分，其中模型榜单只使用通用Harness，不设固定阶段、不挂任何外部工具与知识库，更加纯粹地反映安全智能水平。阿里巴巴此次登顶的正是模型榜单。\n在最新发布的CyberGym模型榜单上，阿里巴巴伏渊息壤大模型（XekRung-27B）以88.9%的成功率排名榜首；排名第2到第5的模型分别由谷歌、OpenAI、智谱和DeepSeek研发，它们的成功率分别为86.3%、85.6%、84.5%和83.3%。\n据阿里巴巴人工智能治理研究中心官方账号透露，伏渊息壤大模型基于阿里自研Qwen3.8-27B多模态大模型，经安全领域训练与极致智能压缩，27B的参数量仅为GPT-5.5-Cyber、Claude Mythos等同类模型的1/27~1/370。这意味着伏渊息壤大模型“既小又强”，实现了极致性价比，用其检测安全漏洞，可以大大节约算力成本，有助于推动安全能力的普惠。\n阿里巴巴集团安全AGI实验室负责人黄龙涛表示，AI大大提升了系统安全的检测率和安全漏洞的检出率，已经成为守护数字世界安全的关键力量。基于“安全智能性价比”的基础，伏渊息壤未来将继续在对抗智能、自进化、Agentic等方向迭代升级。他强调，阿里巴巴安全能力坚持“守己”与“利他”并举，最新安全技术并不只为阿里自身业务保驾护航，也将通过多种方式向全社会开放服务。（勃潺）","image_url":"https://n.sinaimg.cn/spider20260928/380/w660h520/20260928/66d9-2e32729d3be8d66e3be7d9ac767d6ac5.jpg","lang":"zh","published_at":"2026-09-28T05:53:42+00:00","fetched_at":"2026-09-28T06:15:08+00:00","status":"read","starred":0,"extract_state":"ok","summary_auto":"阿里巴巴“小强模型”登顶CyberGym模型榜\n（来源：环球网资讯）\n来源：环球网\n【环球网科技综合报道】日前，阿里巴巴伏渊息壤大模型（XekRung-27B）以88.9%的成功率登上CyberGym榜首，成为该国际权威AI安全能力榜单历史上第一个来自中国的冠军模型。伏渊息壤在获得最高成功率的同时，还实现了“既小又强”的突破，27B的参数量仅为其他上榜模型的1/27~1/370，这意味着该“小强模型”极大降低了智能排查安全漏洞的算力成本…","cluster_id":null,"extract_retries":0,"extract_error":null,"contract_version":"news_item.v1","format_contract_version":"news_item_formats.v1","dedup_url":"https://finance.sina.cn/stock/jdts/2026-09-28/detail-initkarm3874841.d.html?oid=800&vt=4&cid=76993&node_id=76993","quality_profile":{"profile_version":"extraction_quality.v2","bucket":"high","confidence":0.9,"failure_kind":"none","retryable":false,"retry_after_attempts":0,"reason":"High confidence: full text extraction produced 905 characters.","operator_guidance":{"severity":"ok","recommended_action":"trust_full_text","next_step":"Use the extracted full text as the primary article source.","operator_label":"Ready","can_retry":false,"can_use_summary":false,"diagnostics_required":false},"content_depth":{"contract_version":"content_depth.v1","category":"full_text","label":"Full text","has_full_text":true,"has_summary":true,"content_length":905,"summary_length":221,"usable_text_length":905,"source_field":"content"},"legacy_collapsed":false,"signals":{"extract_state":"ok","extract_error":null,"extract_retries":0,"content_length":905,"summary_length":221}},"news_item":{"id":91659,"canonical_url":"https://finance.sina.cn/stock/jdts/2026-09-28/detail-initkarm3874841.d.html?oid=800&vt=4&cid=76993&node_id=76993","source_url":"https://finance.sina.cn/stock/jdts/2026-09-28/detail-initkarm3874841.d.html?oid=800&vt=4&cid=76993&node_id=76993","title":"阿里巴巴“小强模型”登顶CyberGym模型榜 - finance.sina.cn","source_name":"finance.sina.cn","author":null,"published_at":"2026-09-28T05:53:42+00:00","locale":"zh","topic":"ai","tags":[],"rss_summary":"<a href=\"https://news.google.com/rss/articles/CBMisgFBVV95cUxQVFJ1S2pIWWNadkVEcHYweUQzdTI4d2ZqZjFoUmxOMXNXd29XWnhmNHV1T2NSQVpKeTdnYjRaeEdHMUNLMDBlcjZMZjZpd3RBSVVMMEdqZ3F1VEJ6TWM1bFZ1clo3OWk3djZONmZ2S2tBSWYxZFJ4dHY1Q1BNdENWZWJ5UF85czVyVnlNWWpfRWxjakxXY2ZXQTFZWHBfaW43NTVscFZwYlduSFduanVQU3Jn?oc=5\" target=\"_blank\">阿里巴巴“小强模型”登顶CyberGym模型榜</a>&nbsp;&nbsp;<font color=\"#6f6f6f\">finance.sina.cn</font>","full_text":"阿里巴巴“小强模型”登顶CyberGym模型榜\n（来源：环球网资讯）\n来源：环球网\n【环球网科技综合报道】日前，阿里巴巴伏渊息壤大模型（XekRung-27B）以88.9%的成功率登上CyberGym榜首，成为该国际权威AI安全能力榜单历史上第一个来自中国的冠军模型。伏渊息壤在获得最高成功率的同时，还实现了“既小又强”的突破，27B的参数量仅为其他上榜模型的1/27~1/370，这意味着该“小强模型”极大降低了智能排查安全漏洞的算力成本。\nCyberGym是目前国际上最权威的AI安全能力榜单之一，由美国加州大学伯克利分校专业研究团队运营，其背后是一套AI安全能力基准测试框架，涵盖188个软件项目的1507个真实安全漏洞。Cybergym榜单分为模型榜单和智能体榜单两部分，其中模型榜单只使用通用Harness，不设固定阶段、不挂任何外部工具与知识库，更加纯粹地反映安全智能水平。阿里巴巴此次登顶的正是模型榜单。\n在最新发布的CyberGym模型榜单上，阿里巴巴伏渊息壤大模型（XekRung-27B）以88.9%的成功率排名榜首；排名第2到第5的模型分别由谷歌、OpenAI、智谱和DeepSeek研发，它们的成功率分别为86.3%、85.6%、84.5%和83.3%。\n据阿里巴巴人工智能治理研究中心官方账号透露，伏渊息壤大模型基于阿里自研Qwen3.8-27B多模态大模型，经安全领域训练与极致智能压缩，27B的参数量仅为GPT-5.5-Cyber、Claude Mythos等同类模型的1/27~1/370。这意味着伏渊息壤大模型“既小又强”，实现了极致性价比，用其检测安全漏洞，可以大大节约算力成本，有助于推动安全能力的普惠。\n阿里巴巴集团安全AGI实验室负责人黄龙涛表示，AI大大提升了系统安全的检测率和安全漏洞的检出率，已经成为守护数字世界安全的关键力量。基于“安全智能性价比”的基础，伏渊息壤未来将继续在对抗智能、自进化、Agentic等方向迭代升级。他强调，阿里巴巴安全能力坚持“守己”与“利他”并举，最新安全技术并不只为阿里自身业务保驾护航，也将通过多种方式向全社会开放服务。（勃潺）","excerpt":"阿里巴巴“小强模型”登顶CyberGym模型榜\n（来源：环球网资讯）\n来源：环球网\n【环球网科技综合报道】日前，阿里巴巴伏渊息壤大模型（XekRung-27B）以88.9%的成功率登上CyberGym榜首，成为该国际权威AI安全能力榜单历史上第一个来自中国的冠军模型。伏渊息壤在获得最高成功率的同时，还实现了“既小又强”的突破，27B的参数量仅为其他上榜模型的1/27~1/370，这意味着该“小强模型”极大降低了智能排查安全漏洞的算力成本…","extraction":{"state":"ok","confidence":0.9,"error":null,"explanation":"High confidence: full text extraction produced 905 characters.","diagnostics_url":"/api/diagnose?url=https%3A//finance.sina.cn/stock/jdts/2026-09-28/detail-initkarm3874841.d.html%3Foid%3D800%26vt%3D4%26cid%3D76993%26node_id%3D76993","quality_profile":{"profile_version":"extraction_quality.v2","bucket":"high","confidence":0.9,"failure_kind":"none","retryable":false,"retry_after_attempts":0,"reason":"High confidence: full text extraction produced 905 characters.","operator_guidance":{"severity":"ok","recommended_action":"trust_full_text","next_step":"Use the extracted full text as the primary article source.","operator_label":"Ready","can_retry":false,"can_use_summary":false,"diagnostics_required":false},"content_depth":{"contract_version":"content_depth.v1","category":"full_text","label":"Full text","has_full_text":true,"has_summary":true,"content_length":905,"summary_length":221,"usable_text_length":905,"source_field":"content"},"legacy_collapsed":false,"signals":{"extract_state":"ok","extract_error":null,"extract_retries":0,"content_length":905,"summary_length":221}}},"display_formats":["compact","card","full","digest_section","json"]},"daily_stack_record":{"title":"阿里巴巴“小强模型”登顶CyberGym模型榜 - finance.sina.cn","url":"https://finance.sina.cn/stock/jdts/2026-09-28/detail-initkarm3874841.d.html?oid=800&vt=4&cid=76993&node_id=76993","summary":"阿里巴巴“小强模型”登顶CyberGym模型榜\n（来源：环球网资讯）\n来源：环球网\n【环球网科技综合报道】日前，阿里巴巴伏渊息壤大模型（XekRung-27B）以88.9%的成功率登上CyberGym榜首，成为该国际权威AI安全能力榜单历史上第一个来自中国的冠军模型。伏渊息壤在获得最高成功率的同时，还实现了“既小又强”的突破，27B的参数量仅为其他上榜模型的1/27~1/370，这意味着该“小强模型”极大降低了智能排查安全漏洞的算力成本…","source":"finance.sina.cn","date":"2026-09-28T05:53:42+00:00","content":"阿里巴巴“小强模型”登顶CyberGym模型榜\n（来源：环球网资讯）\n来源：环球网\n【环球网科技综合报道】日前，阿里巴巴伏渊息壤大模型（XekRung-27B）以88.9%的成功率登上CyberGym榜首，成为该国际权威AI安全能力榜单历史上第一个来自中国的冠军模型。伏渊息壤在获得最高成功率的同时，还实现了“既小又强”的突破，27B的参数量仅为其他上榜模型的1/27~1/370，这意味着该“小强模型”极大降低了智能排查安全漏洞的算力成本。\nCyberGym是目前国际上最权威的AI安全能力榜单之一，由美国加州大学伯克利分校专业研究团队运营，其背后是一套AI安全能力基准测试框架，涵盖188个软件项目的1507个真实安全漏洞。Cybergym榜单分为模型榜单和智能体榜单两部分，其中模型榜单只使用通用Harness，不设固定阶段、不挂任何外部工具与知识库，更加纯粹地反映安全智能水平。阿里巴巴此次登顶的正是模型榜单。\n在最新发布的CyberGym模型榜单上，阿里巴巴伏渊息壤大模型（XekRung-27B）以88.9%的成功率排名榜首；排名第2到第5的模型分别由谷歌、OpenAI、智谱和DeepSeek研发，它们的成功率分别为86.3%、85.6%、84.5%和83.3%。\n据阿里巴巴人工智能治理研究中心官方账号透露，伏渊息壤大模型基于阿里自研Qwen3.8-27B多模态大模型，经安全领域训练与极致智能压缩，27B的参数量仅为GPT-5.5-Cyber、Claude Mythos等同类模型的1/27~1/370。这意味着伏渊息壤大模型“既小又强”，实现了极致性价比，用其检测安全漏洞，可以大大节约算力成本，有助于推动安全能力的普惠。\n阿里巴巴集团安全AGI实验室负责人黄龙涛表示，AI大大提升了系统安全的检测率和安全漏洞的检出率，已经成为守护数字世界安全的关键力量。基于“安全智能性价比”的基础，伏渊息壤未来将继续在对抗智能、自进化、Agentic等方向迭代升级。他强调，阿里巴巴安全能力坚持“守己”与“利他”并举，最新安全技术并不只为阿里自身业务保驾护航，也将通过多种方式向全社会开放服务。（勃潺）","confidence":0.9,"diagnostics_url":"/api/diagnose?url=https%3A//finance.sina.cn/stock/jdts/2026-09-28/detail-initkarm3874841.d.html%3Foid%3D800%26vt%3D4%26cid%3D76993%26node_id%3D76993","quality_bucket":"high","failure_kind":"none","retryable":false,"quality_reason":"High confidence: full text extraction produced 905 characters.","quality_profile":{"profile_version":"extraction_quality.v2","bucket":"high","confidence":0.9,"failure_kind":"none","retryable":false,"retry_after_attempts":0,"reason":"High confidence: full text extraction produced 905 characters.","operator_guidance":{"severity":"ok","recommended_action":"trust_full_text","next_step":"Use the extracted full text as the primary article source.","operator_label":"Ready","can_retry":false,"can_use_summary":false,"diagnostics_required":false},"content_depth":{"contract_version":"content_depth.v1","category":"full_text","label":"Full text","has_full_text":true,"has_summary":true,"content_length":905,"summary_length":221,"usable_text_length":905,"source_field":"content"},"legacy_collapsed":false,"signals":{"extract_state":"ok","extract_error":null,"extract_retries":0,"content_length":905,"summary_length":221}},"tags":[]},"fallback_formats":["markdown","json","html"],"actions":{"read":"/item/91659","export_markdown":"/api/items/91659/export?format=markdown","export_json":"/api/items/91659/export?format=json","diagnose":"/api/diagnose?url=https%3A//finance.sina.cn/stock/jdts/2026-09-28/detail-initkarm3874841.d.html%3Foid%3D800%26vt%3D4%26cid%3D76993%26node_id%3D76993"},"formats":{"full":{"id":91659,"title":"阿里巴巴“小强模型”登顶CyberGym模型榜 - finance.sina.cn","url":"https://finance.sina.cn/stock/jdts/2026-09-28/detail-initkarm3874841.d.html?oid=800&vt=4&cid=76993&node_id=76993","source":"finance.sina.cn","author":null,"published_at":"2026-09-28T05:53:42+00:00","locale":"zh","topic":"ai","tags":[],"excerpt":"阿里巴巴“小强模型”登顶CyberGym模型榜\n（来源：环球网资讯）\n来源：环球网\n【环球网科技综合报道】日前，阿里巴巴伏渊息壤大模型（XekRung-27B）以88.9%的成功率登上CyberGym榜首，成为该国际权威AI安全能力榜单历史上第一个来自中国的冠军模型。伏渊息壤在获得最高成功率的同时，还实现了“既小又强”的突破，27B的参数量仅为其他上榜模型的1/27~1/370，这意味着该“小强模型”极大降低了智能排查安全漏洞的算力成本…","full_text":"阿里巴巴“小强模型”登顶CyberGym模型榜\n（来源：环球网资讯）\n来源：环球网\n【环球网科技综合报道】日前，阿里巴巴伏渊息壤大模型（XekRung-27B）以88.9%的成功率登上CyberGym榜首，成为该国际权威AI安全能力榜单历史上第一个来自中国的冠军模型。伏渊息壤在获得最高成功率的同时，还实现了“既小又强”的突破，27B的参数量仅为其他上榜模型的1/27~1/370，这意味着该“小强模型”极大降低了智能排查安全漏洞的算力成本。\nCyberGym是目前国际上最权威的AI安全能力榜单之一，由美国加州大学伯克利分校专业研究团队运营，其背后是一套AI安全能力基准测试框架，涵盖188个软件项目的1507个真实安全漏洞。Cybergym榜单分为模型榜单和智能体榜单两部分，其中模型榜单只使用通用Harness，不设固定阶段、不挂任何外部工具与知识库，更加纯粹地反映安全智能水平。阿里巴巴此次登顶的正是模型榜单。\n在最新发布的CyberGym模型榜单上，阿里巴巴伏渊息壤大模型（XekRung-27B）以88.9%的成功率排名榜首；排名第2到第5的模型分别由谷歌、OpenAI、智谱和DeepSeek研发，它们的成功率分别为86.3%、85.6%、84.5%和83.3%。\n据阿里巴巴人工智能治理研究中心官方账号透露，伏渊息壤大模型基于阿里自研Qwen3.8-27B多模态大模型，经安全领域训练与极致智能压缩，27B的参数量仅为GPT-5.5-Cyber、Claude Mythos等同类模型的1/27~1/370。这意味着伏渊息壤大模型“既小又强”，实现了极致性价比，用其检测安全漏洞，可以大大节约算力成本，有助于推动安全能力的普惠。\n阿里巴巴集团安全AGI实验室负责人黄龙涛表示，AI大大提升了系统安全的检测率和安全漏洞的检出率，已经成为守护数字世界安全的关键力量。基于“安全智能性价比”的基础，伏渊息壤未来将继续在对抗智能、自进化、Agentic等方向迭代升级。他强调，阿里巴巴安全能力坚持“守己”与“利他”并举，最新安全技术并不只为阿里自身业务保驾护航，也将通过多种方式向全社会开放服务。（勃潺）","reading_time_min":1,"extraction":{"state":"ok","confidence":0.9,"error":null,"explanation":"High confidence: full text extraction produced 905 characters.","diagnostics_url":"/api/diagnose?url=https%3A//finance.sina.cn/stock/jdts/2026-09-28/detail-initkarm3874841.d.html%3Foid%3D800%26vt%3D4%26cid%3D76993%26node_id%3D76993","quality_profile":{"profile_version":"extraction_quality.v2","bucket":"high","confidence":0.9,"failure_kind":"none","retryable":false,"retry_after_attempts":0,"reason":"High confidence: full text extraction produced 905 characters.","operator_guidance":{"severity":"ok","recommended_action":"trust_full_text","next_step":"Use the extracted full text as the primary article source.","operator_label":"Ready","can_retry":false,"can_use_summary":false,"diagnostics_required":false},"content_depth":{"contract_version":"content_depth.v1","category":"full_text","label":"Full text","has_full_text":true,"has_summary":true,"content_length":905,"summary_length":221,"usable_text_length":905,"source_field":"content"},"legacy_collapsed":false,"signals":{"extract_state":"ok","extract_error":null,"extract_retries":0,"content_length":905,"summary_length":221}}},"quality_profile":{"profile_version":"extraction_quality.v2","bucket":"high","confidence":0.9,"failure_kind":"none","retryable":false,"retry_after_attempts":0,"reason":"High confidence: full text extraction produced 905 characters.","operator_guidance":{"severity":"ok","recommended_action":"trust_full_text","next_step":"Use the extracted full text as the primary article source.","operator_label":"Ready","can_retry":false,"can_use_summary":false,"diagnostics_required":false},"content_depth":{"contract_version":"content_depth.v1","category":"full_text","label":"Full text","has_full_text":true,"has_summary":true,"content_length":905,"summary_length":221,"usable_text_length":905,"source_field":"content"},"legacy_collapsed":false,"signals":{"extract_state":"ok","extract_error":null,"extract_retries":0,"content_length":905,"summary_length":221}},"actions":{"read":"/item/91659","export_markdown":"/api/items/91659/export?format=markdown","export_json":"/api/items/91659/export?format=json","diagnose":"/api/diagnose?url=https%3A//finance.sina.cn/stock/jdts/2026-09-28/detail-initkarm3874841.d.html%3Foid%3D800%26vt%3D4%26cid%3D76993%26node_id%3D76993"}},"digest":{"id":91659,"title":"阿里巴巴“小强模型”登顶CyberGym模型榜 - finance.sina.cn","url":"https://finance.sina.cn/stock/jdts/2026-09-28/detail-initkarm3874841.d.html?oid=800&vt=4&cid=76993&node_id=76993","source":"finance.sina.cn","topic":"ai","published_at":"2026-09-28T05:53:42+00:00","excerpt":"阿里巴巴“小强模型”登顶CyberGym模型榜 （来源：环球网资讯） 来源：环球网 【环球网科技综合报道】日前，阿里巴巴伏渊息壤大模型（XekRung-27B）以88.9%的成功率登上CyberGym榜首，成为该国际权威AI安全能力榜单历史上第一个来自中国的冠军模型。伏渊息壤在获得最高成功率的同时，还实现了“既小又强”的突破，27B的参数量仅为其他上榜模型的1/27~1/370，这意味着该“小强模型”极大降低了智能排查安全漏洞的算力成本…","quality_bucket":"high","quality_reason":"High confidence: full text extraction produced 905 characters.","reading_time_min":1,"cluster_id":null},"card":{"display_title":"阿里巴巴“小强模型”登顶CyberGym模型榜 - finance.sina.cn","subtitle":"finance.sina.cn · 2026-09-28","summary":"阿里巴巴“小强模型”登顶CyberGym模型榜 （来源：环球网资讯） 来源：环球网…","badges":["quality:high"],"links":{"read":"/item/91659","original":"https://finance.sina.cn/stock/jdts/2026-09-28/detail-initkarm3874841.d.html?oid=800&vt=4&cid=76993&node_id=76993","diagnose":"/api/diagnose?url=https%3A//finance.sina.cn/stock/jdts/2026-09-28/detail-initkarm3874841.d.html%3Foid%3D800%26vt%3D4%26cid%3D76993%26node_id%3D76993"},"quality_warning":null},"export":{"title":"阿里巴巴“小强模型”登顶CyberGym模型榜 - finance.sina.cn","url":"https://finance.sina.cn/stock/jdts/2026-09-28/detail-initkarm3874841.d.html?oid=800&vt=4&cid=76993&node_id=76993","summary":"阿里巴巴“小强模型”登顶CyberGym模型榜\n（来源：环球网资讯）\n来源：环球网\n【环球网科技综合报道】日前，阿里巴巴伏渊息壤大模型（XekRung-27B）以88.9%的成功率登上CyberGym榜首，成为该国际权威AI安全能力榜单历史上第一个来自中国的冠军模型。伏渊息壤在获得最高成功率的同时，还实现了“既小又强”的突破，27B的参数量仅为其他上榜模型的1/27~1/370，这意味着该“小强模型”极大降低了智能排查安全漏洞的算力成本…","source":"finance.sina.cn","date":"2026-09-28T05:53:42+00:00","content":"阿里巴巴“小强模型”登顶CyberGym模型榜\n（来源：环球网资讯）\n来源：环球网\n【环球网科技综合报道】日前，阿里巴巴伏渊息壤大模型（XekRung-27B）以88.9%的成功率登上CyberGym榜首，成为该国际权威AI安全能力榜单历史上第一个来自中国的冠军模型。伏渊息壤在获得最高成功率的同时，还实现了“既小又强”的突破，27B的参数量仅为其他上榜模型的1/27~1/370，这意味着该“小强模型”极大降低了智能排查安全漏洞的算力成本。\nCyberGym是目前国际上最权威的AI安全能力榜单之一，由美国加州大学伯克利分校专业研究团队运营，其背后是一套AI安全能力基准测试框架，涵盖188个软件项目的1507个真实安全漏洞。Cybergym榜单分为模型榜单和智能体榜单两部分，其中模型榜单只使用通用Harness，不设固定阶段、不挂任何外部工具与知识库，更加纯粹地反映安全智能水平。阿里巴巴此次登顶的正是模型榜单。\n在最新发布的CyberGym模型榜单上，阿里巴巴伏渊息壤大模型（XekRung-27B）以88.9%的成功率排名榜首；排名第2到第5的模型分别由谷歌、OpenAI、智谱和DeepSeek研发，它们的成功率分别为86.3%、85.6%、84.5%和83.3%。\n据阿里巴巴人工智能治理研究中心官方账号透露，伏渊息壤大模型基于阿里自研Qwen3.8-27B多模态大模型，经安全领域训练与极致智能压缩，27B的参数量仅为GPT-5.5-Cyber、Claude Mythos等同类模型的1/27~1/370。这意味着伏渊息壤大模型“既小又强”，实现了极致性价比，用其检测安全漏洞，可以大大节约算力成本，有助于推动安全能力的普惠。\n阿里巴巴集团安全AGI实验室负责人黄龙涛表示，AI大大提升了系统安全的检测率和安全漏洞的检出率，已经成为守护数字世界安全的关键力量。基于“安全智能性价比”的基础，伏渊息壤未来将继续在对抗智能、自进化、Agentic等方向迭代升级。他强调，阿里巴巴安全能力坚持“守己”与“利他”并举，最新安全技术并不只为阿里自身业务保驾护航，也将通过多种方式向全社会开放服务。（勃潺）","confidence":0.9,"diagnostics_url":"/api/diagnose?url=https%3A//finance.sina.cn/stock/jdts/2026-09-28/detail-initkarm3874841.d.html%3Foid%3D800%26vt%3D4%26cid%3D76993%26node_id%3D76993","quality_bucket":"high","failure_kind":"none","retryable":false,"quality_reason":"High confidence: full text extraction produced 905 characters.","quality_profile":{"profile_version":"extraction_quality.v2","bucket":"high","confidence":0.9,"failure_kind":"none","retryable":false,"retry_after_attempts":0,"reason":"High confidence: full text extraction produced 905 characters.","operator_guidance":{"severity":"ok","recommended_action":"trust_full_text","next_step":"Use the extracted full text as the primary article source.","operator_label":"Ready","can_retry":false,"can_use_summary":false,"diagnostics_required":false},"content_depth":{"contract_version":"content_depth.v1","category":"full_text","label":"Full text","has_full_text":true,"has_summary":true,"content_length":905,"summary_length":221,"usable_text_length":905,"source_field":"content"},"legacy_collapsed":false,"signals":{"extract_state":"ok","extract_error":null,"extract_retries":0,"content_length":905,"summary_length":221}},"tags":[],"format_contract_version":"news_item_formats.v1"}}}