{"schema":"hottub-news/story/v1","story":"st_c6d3katt563o4bpqcvgq","title":"Google最強AI模型Gemini 4 Argon登場 長程式開發測試擊敗Claude","title_en":"Google's strongest AI model Gemini 4 Argon debuts, long-term code development and testing surpass Claude","outlets":2,"brief":null,"page":"https://hottub.news/stories/googles-cutting-edge-ai-model-gemini-4-argon-debuts-long-term-code-c6d3katt56","count":2,"items":[{"id":"n_espv6wwlgrkjdmsnjywa","seq":1035460,"kind":"article","title":"Google最強AI模型Gemini 4 Argon登場 長程式開發測試擊敗Claude","title_en":"Google's strongest AI model Gemini 4 Argon debuts, long-term code development and testing surpass Claude","url":"https://ai.ettoday.net/news/3246882","summary":"Google 最新 AI 模型 Gemini 4 Argon 正式登場，主打長時間軟體工程、企業知識工作與網路安全防禦等複雜任務。Google 公布的測試結果顯示，Gemini 4 Argon 在 DeepSWE v1.1 長周期軟體工程測試取得 77.9%，高於 Claude Opus 5.5 的 74.2% 與 GPT-6 Astra 的 74.1%。 《詳全文...》","published":"2026-10-01T00:56:00Z","seen":"2026-10-01T02:25:17Z","lang":"zh","country":"TW","source":{"id":"kite-feedburner-com-c31c9e","name":"feedburner.com","domain":"feeds.feedburner.com"},"via":"kite-feedburner-com-c31c9e","entities":[{"id":"Q95","name":"Google","type":"org"},{"id":"Q979959","name":"Claude","type":"other"}],"story":"st_c6d3katt563o4bpqcvgq","category":"science-technology","category_p":0.8799999952316284,"sentiment":"neutral","tone":0.25999999046325684,"political":0.05000000074505806},{"id":"n_c6d3katt563o4bpqcvgq","seq":1023625,"kind":"article","title":"谷歌最前沿 AI 模型：Gemini 4 Argon 登场，长周期代码工程 DeepSWE 测试超 Claude Opus 5.5","title_en":"Google's cutting-edge AI model: Gemini 4 Argon debuts, long-term code engineering DeepSWE test surpasses Claude opus 5.5","url":"https://www.ithome.com/1/008/953.htm","summary":"IT之家 10 月 1 日消息，谷歌昨日（9 月 30 日）发布 Gemini 4 Argon， 称其为公司迄今最先进的 AI 模型，但目前并未面向公众全面开放。 定位方面，Gemini 4 Argon 是谷歌当前“最先进”模型，重点是长流程软件工程、企业知识工作和网络安全防御。官方称其在真实世界软件工程测试创纪录，网络安全基准并列第一，并在金融、法律等专业任务基准领先。 模型特性方面，Gemini 4 Argon 单次输出上限达到约 100 万 tokens，高于此前的 64,000 tokens，适合处理大型代码库、长文档和多阶段任务。 谷歌 DeepMind Gemini 产品负责人 Tulsee Doshi 表示：“Argon 是一个功能全面的模型，在多个领域都具备前沿能力。” 在新模型在某些编码和知识工作基准测试中处于最先进水平，在多个方面都优于竞争对手 OpenAI 的前沿模型 GPT-6 Astra。 性能方面，谷歌宣称其 DeepSWE v1.1 得分为 77.9%。Claude Opus 5.5 的得分为…","published":"2026-09-30T22:57:36Z","seen":"2026-10-01T00:10:41Z","lang":"zh","country":"CN","source":{"id":"kite-ithome-com-f1cdd0","name":"ithome.com","domain":"ithome.com","outlet":{"wikidata":"Q28413779","name":"IThome","type":"website","country":"CN","designations":["website"]}},"via":"kite-ithome-com-f1cdd0","entities":[{"id":"Q95","name":"Google","type":"org"}],"story":"st_c6d3katt563o4bpqcvgq","category":"science-technology","category_p":0.9700000286102295,"sentiment":"neutral","tone":0.17000000178813934,"political":0.03999999910593033}]}