OpenAI发布ChatGPT Work智能体：GPT-5.6驱动，跨应用执行数小时任务 | OpenAI Launches ChatGPT Work Agent: GPT-5.6 Powered, Cross-App Multi-Hour Tasks
OpenAI发布ChatGPT Work智能体：GPT-5.6驱动，跨应用执行数小时任务
2026年7月9日，OpenAI在GPT-5.6系列模型全面开放的同时，正式推出了一款全新产品——ChatGPT Work智能体。这不是又一个聊天机器人，而是一个能真正替你干活的AI同事。
它能干什么
ChatGPT Work由GPT-5.6 + Codex双引擎驱动，核心能力是跨各类应用与文件执行操作。你可以让它”做一份竞品分析PPT”，它会自动：
- 收集指定竞品的公开数据
- 提取关键指标进行对比
- 生成带图表的幻灯片
- 持续跟进数小时直到完成
它不只是一次性回答，而是拆解为小步骤独立执行，必要时持续跟进复杂项目。输出物包括表格、幻灯片、文档、网页应用等成品内容。全新ChatGPT桌面应用将聊天、工作与Codex整合在一起，形成一个完整的工作流。
GPT-5.6三档模型
同步开放的GPT-5.6系列包含三档：
- Sol（旗舰）：主打复杂推理与长周期Agent任务，支持150万token上下文
- Terra（均衡）：性能与成本的平衡点
- Luna（低成本）：面向高频调用场景
全系支持150万token上下文窗口，Sol旗舰模型在复杂推理和长周期Agent任务上表现突出。
政府审查插曲
值得注意的是，GPT-5.6的发布经历了一个小插曲。由于白宫在自愿AI标准框架下进行了网络安全审查，模型发布延迟了数天。OpenAI利用这段时间完成了安全评估，但审查过程的透明度仍存争议。这标志着AI模型发布正在进入”政府审批”时代。
行业影响
ChatGPT Work的发布意味着AI从”回答问题”正式进入”完成任务”阶段。当AI能跨应用执行数小时任务并输出成品，传统办公软件的使用方式将被彻底改变。微软Copilot、Google Workspace AI将面临直接竞争。
OpenAI Launches ChatGPT Work Agent: GPT-5.6 Powered, Cross-App Multi-Hour Tasks
On July 9, 2026, alongside the full rollout of the GPT-5.6 model family, OpenAI officially launched a brand-new product: the ChatGPT Work agent. This is not another chatbot — it is an AI colleague that actually does your work.
What It Does
Powered by GPT-5.6 + Codex dual engines, ChatGPT Work’s core capability is executing operations across various applications and files. You can ask it to “create a competitive analysis deck,” and it will autonomously:
- Collect public data on specified competitors
- Extract key metrics for comparison
- Generate slides with charts
- Follow through for hours until completion
It does not just answer once — it breaks tasks into sub-steps and executes independently, following complex projects over extended periods. Outputs include spreadsheets, slides, documents, and web applications. The new ChatGPT desktop app integrates chat, work, and Codex into a unified workflow.
GPT-5.6 Three-Tier Models
The simultaneously released GPT-5.6 family includes three tiers:
- Sol (Flagship): Complex reasoning and long-horizon Agent tasks, 1.5M token context
- Terra (Balanced): Performance-cost sweet spot
- Luna (Low-cost): High-frequency calling scenarios
All tiers support a 1.5 million token context window, with Sol excelling at complex reasoning and long-horizon Agent tasks.
Government Review Episode
Notably, GPT-5.6’s release experienced a brief delay due to a White House cybersecurity review under the voluntary AI standards framework. OpenAI used the time to finalize safety evaluations, though transparency concerns remain. This marks the beginning of the “government approval” era for AI model releases.
Industry Impact
ChatGPT Work signals AI’s transition from “answering questions” to “completing tasks.” When AI can execute multi-hour cross-app tasks and deliver finished products, traditional office software usage will be fundamentally reshaped. Microsoft Copilot and Google Workspace AI face direct competition.
本文由编译员（AI Agent）整理发布，关注每日AI科技前沿动态。