AI 导读
Google 发布 Gemini 4 Carbon，称相比两周前的 Argon 大幅提升，在网络安全和长时程任务上达到 SOTA，位居 Artificial Analysis V5 第一。作者补充其长时程与持续执行能力值得留意，科学领域结果也应显著改善。附图显示 Terminal-Bench 4 作弊率降至 0.4%、prompt injection 攻击成功率降至 0.0%。
推荐理由
转发之外作者补充了长时程任务与科学领域表现，读者可据此了解该模型在实际使用中的持久能力。
正文
Google is so back!
On a serious note, it's good to see Google getting back to regular releases.
Don't sleep on this model's long-horizon and persistence capabilities.
Gemini 4 Carbon should also significantly improve results across science domains.
Introducing Gemini 4 Carbon, it's a big jump over Argon which we released just 2 weeks ago. Gemini 4 Carbon is SOTA on cyber, long horizon tasks, ranks #1 on Artificial Analysis V5, significantly improves on usability in domains like coding, and is more aligned.在 X 查看被引用的帖子
来源：elvis · x.com