Gemini 4 Carbon 发布:编程能力大幅提升Gemini 4 Carbon Drops: Big Coding Jump
Google VP Josh Woodward 宣布 Gemini 4 Carbon 已向全球 Ultra 订阅用户推送。Every CEO Dan Shipper 的 day-zero 实测显示,Carbon 在 Senior Engineer 基准上从 Argon 的 21% 提升到 44%,能长时间独立推进大型项目,3D 建模、视频剪辑和设计细节也有明显进步,但写作能力与 Argon 持平,且偶尔会过度执行任务。Shipper 建议用它做有明确终点的大型任务,并作为第二意见。
Google VP Josh Woodward announced Gemini 4 Carbon is live for Ultra subscribers worldwide. Every CEO Dan Shipper’s day-zero vibe check found Carbon scoring 44% on their Senior Engineer benchmark versus Argon’s 21%. It can run ambitious multi-hour projects, excel at 3D modeling from scans, audio/video editing, and detailed design, but writing is roughly even with Argon and it sometimes over-executes. Shipper recommends clear finish-line tasks and using it as a second opinion.
查看原文 →查看原文 →
Claude 新增 Whiteboards、Sheets、Data 与交互式连接器Claude Rolls Out Whiteboards, Sheets, Data & Interactive Connectors
Anthropic 的 Claude 正式推出 Claude Data(原 Dashboards)、Whiteboards、Sheets,以及可在对话中直接打开 Adobe、Atlassian、Box 完整界面的交互式连接器。Box CEO Aaron Levie 同步宣布与 Anthropic 合作,通过 MCP 实现 Claude 对 Box 内容的完整读写,agent 可直接创建、编辑合同与策略文档。Thibault Sottiaux 称赞这是实验室之间真正的协作创新。
Claude launched Whiteboards, Sheets, Claude Data (formerly Dashboards), and interactive connectors that open the full Adobe, Atlassian or Box UI inside the chat. Box CEO Aaron Levie announced full read/write interoperability with Claude via MCP so agents can create and edit contracts, decks and strategy docs directly. OpenAI’s Thibault Sottiaux called it labs innovating together.
查看原文 →查看原文 →查看原文 →查看原文 →
Aaron Levie:Agent Swarms 将带来下一次生产力跃迁Aaron Levie on Agent Swarms as the Next Productivity Leap
Box CEO Aaron Levie 认为,从聊天到单 agent,再到目标驱动的 agent swarm,生产力上限正在被打破。任何在数字环境中可完成的研究、决策、数据处理或代码生成工作,swarm 都能比传统方式更快完成。游戏开发、交易、生命科学、网络安全等领域将首先受益,相关 harness 与基础设施存在巨大机会。
Box CEO Aaron Levie argues that agent swarms—goal-seeking collectives with no inherent productivity ceiling—will separate teams that adopt them from those that don’t. Any digital work involving research, decisions, data processing or code generation can be completed faster by swarms. Game development, trading, life sciences, cyber and math breakthroughs are early targets; harnesses and infrastructure for this pattern are wide open.
查看原文 →