OpenAI 推出 dots:永遠在線的 GPT-6 Astra 代理,在自己的雲端電腦上運作

2026年9月29日 18:30
站內 AI 整理稿

OpenAI just introduced dots at their DevDay today.Dots are persistent AI agents powered by GPT-6 Astra.Each dot gets its own cloud computer and browser.It works across 4,000+ apps through ChatGPT plugins and keeps going after you log off.Is it deployable today?Yes, as a managed product.

Dots are rolling out in ChatGPT for Pro and Business Premium users in eligible markets.Enterprise, Edu and Healthcare workspaces get a beta once an admin turns it on.There is no self-hosted or open-weights option.

What is a dot A dot turns ChatGPT from a request and response loop into a standing assignment.You give it a goal, a name and boundaries.It then pursues that goal around the clock and learns from your feedback.Each user starts with 1 primary dot.

To be honest, much of this was already possible with Codex and similar agent harnesses.The change here is packaging: 1 persistent agent, 1 identity, 1 dedicated machine.How a dot works Own computer: Each dot runs on a separate cloud computer with a browser.

Your laptop stays isolated unless you connect it.You can open the dot’s computer at any time to inspect its work.Tools: It uses the plugins you connect across 4,000+ apps.It can also start tasks in Codex and ChatGPT Work for research, analysis, documents and software.

Surfaces: You can message or call your dot in ChatGPT on web, desktop and mobile.Slack and Teams are supported, and texting is coming soon.Proactive research: When you are not working with it, the dot looks for ways to help.

In this background mode, connected apps are read-only, so it cannot send messages or change content.The model underneath is also important to know.In OpenAI’s latency simulations, GPT-6 Astra scores 72.6% on OSWorld 2.0 at roughly 40 minutes per task.GPT-5.6 Sol scored 65.7% at roughly 75 minutes.

Control and safety layer Built-in rules decide when a dot acts alone and when it asks for approval.Custom Rules let you allow, block or require approval for specific actions.Auto-review checks any action that could affect your accounts or share information.

Activity View shows progress, including background work, so you can redirect the dot.Credential handling: Dots sign in with saved passwords without exposing them to the model.Sensitive tasks, such as changing a password, stay with you.

Monitoring can pause or stop a dot if it detects a safety concern.OpenAI details this in a separate safety post.It also states that dots can still make mistakes, so consequential work needs review.The caution is timely.

A day before launch, OpenAI team stated that its agents had posted users’ images online, affecting 53 ChatGPT users.Specialist dots for organizations OpenAI is also previewing specialist dots.Each gets its own identity, credentials and access to company systems for a fixed role.

OpenAI tested them internally across procurement, invoice processing, email marketing, customer support and commercial contracting.Rollout starts with enterprise pilots.OpenAI is working with Microsoft to bring specialist dots to Agent 365.

Pricing and usage limits The first dot is included in Pro and Business Premium plans at no extra cost.Conversations with a dot do not count toward ChatGPT usage limits.Tasks it starts in Codex or ChatGPT Work do count.OpenAI also added a $500 paid tier the same day.

Check the ChatGPT pricing page for current plan limits.

Dots vs Meta Muse vs Instinct: Feature Comparison FeatureOpenAI dotsMeta MuseInstinctLaunchSep 29, 2026Sep 8, 2026Invite-only since Aug 2026Underlying modelGPT-6 AstraMuse SparkNot disclosedWhere the agent runsOwn cloud computer and browserMuse Secure VM, a dedicated cloud computerOwn computer and phone numberAction gateCustom Rules, auto-review, safety monitorSeparate Sentinel agent approves outbound actionsNo public details foundInterfacesChatGPT web, desktop, mobile; Slack; Teams; texting sooniOS, Android, muse.

ai, WhatsAppSMS and texting; no mobile app yetStandout featureRead-only proactive research; 4,000+ appsDeep ties to Meta appsConcierge phone calls; agent-to-agent plans with friendsOrg or team agentsSpecialist dots; Agent 365 work with MicrosoftConsumer launch; Axios reports a shift toward work useConsumer-focused personal assistantAvailabilityPro and Business Premium, eligible marketsUS onlyInvite-onlyPriceIncluded in Pro and Business PremiumFree, $20 or $100 per monthNo public pricing found The split is clear.

Muse chases mass consumer reach.Instinct lives in text messages.Dots start with high-end subscribers, mostly developers.Interactive explainer: inside a dot’s work loop

Goal Cloud computer Apps via plugins Auto-review Outcome Choose a goal and a rule, then press Run. ApproveDecline

Illustrative simulation of the flow OpenAI describes.Task examples come from OpenAI's launch materials.<!-- PANEL 2 --> You're working with itYou're away Read new email Send a reply Edit a shared doc In background "proactive research," OpenAI says connected apps are read-only.

The dot can look, but it cannot send messages or change content.<!-- PANEL 3 --> Your computer <div cla

Related

相關文章

IT之家生成式AI

報道稱 OpenAI 年化經常性收入接近 700 億美元,較三季度初增長超 70%

作者:潞源 責編:潞源 評論: 9 月 29 日消息,據外媒 Axios 今天報道,據知情人士透露,OpenAI 年化經常性收入(ARR)已接近 700 億美元(注:現匯率約合 4,705.11 億元人民幣)。據報道,自 2026 年三季度初以來,OpenAI 的年化經常性收入增長超 70%,其 B2B(企業對企業)收入增長超 100%。

剛剛
IT之家生成式AI

奔赴南海科考,我國自主研發 4500 米級深海作業機器人啟航

作者:浩渺 責編:浩渺 評論: 感謝網友 很宅很怕生 的線索投遞!9 月 29 日消息,據央視新聞報道,9 月 28 日,在山東威海港,我國自主研發的 150 馬力 / 250 馬力 4500 米級深海作業機器人正式啟航,聯合哈爾濱工業大學(威海)奔赴南海開展聯合科考。

剛剛
鈦媒體生成式AI

ARR真的是大模型企業的黃金指標嗎?

數智奔流2026.09.29 19:13 · 來自河南全文4129字00:00 / 13:08快速上漲的ARR,要花多少錢才能維持。文 | 數智奔流,作者 | 熵野,編輯 | 沈校9月,幾件看似沒有關聯的事件,揭露了大模型公司正在發生的變化。

9 分鐘前
鈦媒體生成式AI

82億美元,硅谷上演頂級版“Girls help girls”

山農下山2026.09.29 18:12 · 來自甘肅全文4164字蘇姿豐與李飛飛“會師”。文 | 山農下山溫柔的收購與現實的考量“去尋找一個更新的世界,永遠不會太晚。”AMD在9月28日官宣收購李飛飛創辦的 World Labs,82億美元、全股票交易。

1 小時前