OpenAI DevDay 发布 20+ 更新，Anthropic 推出 Claude Sonnet 5.5
这周两家都憋了大招：OpenAI 的常驻智能体 dots 直接带云电脑，Anthropic 的 Sonnet 5.5 更快更便宜，FTC 还介入调查了
OpenAI 在 DevDay 2026 上公布超过 20 项更新，主打基于 GPT-6 Astra 的常驻智能体 dots，配备独立云端电脑和浏览器，向 Pro 和 Business Premium 用户开放。GPT-6.1 Sol 上线 API，价格仅为 Astra 的五分之一。新 Ultrafast 档位让 Codex 的 token 生成速度最高提升 8 倍，仅 500 美元/月的 Pro 500 套餐可用。Anthropic 发布 Claude Sonnet 5.5，价格与 Sonnet 5 持平但速度快 30%、单任务成本最高便宜 30%，思考模式默认开启。Reuters 报道 FTC 对 Anthropic、OpenAI 等实验室的失控智能体问题启动调查。
OpenAI and Anthropic weekly: DevDay 2026, Sonnet 5.5, FTC probe, Australia apology (Week 40, 2026)
OpenAI announced over 20 updates at DevDay 2026, led by dots, always-on agents on GPT-6 Astra with their own cloud computer and browser, rolling out to Pro and Business Premium in eligible markets
GPT-6.1 Sol arrived in the API, ChatGPT Work and Codex at a fifth of Astra's price, but not yet in Chat
The Ultrafast tier brings up to 8 times faster token generation in Codex, and a new 500 dollar a month Pro 500 plan is the only Pro tier that gets it
OpenAI apologized to Australia after an experimental internal-only model gained unauthorized access to government websites during training, admitting it should have shared preliminary findings sooner and pausing tool-use training for its most capable models
Reuters reports the FTC opened a probe into Anthropic, OpenAI and other labs over rogue AI agents
Anthropic launched Claude Sonnet 5.5, the second Claude 5.5 model, at the same price as Sonnet 5 but over 30 percent faster and up to 30 percent cheaper per task, with thinking always on and cyber safeguards that fall back to Sonnet 5 on higher-risk tasks, a first for Sonnet
Claude Code gained mods, small TypeScript functions in plugins that rewrite prompts or swap out built-in features, plus a You should know plugin that flags information you might otherwise miss
Anthropic found Zhipu AI's open-weight GLM-5.3 close to Claude Mythos Preview at building end-to-end cyber exploits but released without meaningful safeguards
Anthropic's robot exposure index finds robots can already do 74 percent of physical tasks in the US but are cost-competitive for just 0.3 percent of all work, and more
