Everything OpenAI Announced at DevDay 2026
The complete DevDay 2026 field guide: every major OpenAI announcement, what it does, who gets it, and direct links to the official sources.
What this lesson gives you
A complete, source-linked map of OpenAI DevDay 2026—from Dots and GPT-6.1 Sol to Space, Codex Cloud, plugin extensions, MCP Events, team workflows, Pro 500, and the OpenAI Marketplace.
Best for
Anyone who wants to understand what OpenAI actually launched at DevDay 2026, which features are available now, and how the pieces fit together without piecing it together from dozens of posts.
Inside this lesson
- DevDay 2026 moved OpenAI further from one-off chat toward persistent, ongoing agent work.
- ChatGPT Space, Pages, plugins, and event-driven automations create a shared work layer around the models.
- Every major claim in the guide links directly to the official OpenAI, ChatGPT, developer, or AWS source so readers can verify and go deeper.
The full lesson
A complete field guide from The AGI Guide · Updated September 29, 2026
OpenAI DevDay 2026 was not one product launch. It was a map of where OpenAI wants the entire computing stack to go next.
On September 29, 2026, OpenAI held its annual developer conference at Fort Mason in San Francisco. The company described it as its biggest DevDay yet and published more than 20 major announcements across ChatGPT, Codex, models, agents, plugins, developer APIs, collaboration, security, and subscription plans.
If you only saw the clips about Dots or ChatGPT Space, you saw the headline. You did not see the full platform shift.
The larger story is that OpenAI is trying to make ChatGPT a place where work does not merely begin with a prompt. Work can stay alive. Agents can keep going. Teams can share context. Developers can build native interfaces inside ChatGPT. Software can trigger automations when external events happen. Codex can keep running in the cloud. Your ChatGPT plan can follow you into other products. And companies can deploy OpenAI-powered agents inside their own infrastructure.
This guide is a complete handoff. It explains what DevDay is, what OpenAI announced, which features are available now, which are rolling out or coming later, what each feature is actually for, and where to read the official documentation.
Primary source: OpenAI DevDay 2026 Recap
Event page: OpenAI DevDay 2026
---
The short version
If you only remember ten things from DevDay 2026, remember these:
- Dots are always-on agents powered by GPT-6 Astra that get their own cloud computer, browser, and access to connected apps.
- GPT-6.1 Sol brings near-Astra capability to coding, computer use, and professional work at a much lower API price.
- Ultrafast is a premium speed tier for workloads where latency matters.
- ChatGPT Space is a shared workspace where people, ChatGPT, and Dots work from the same context.
- Pages and collaborative slides turn ChatGPT into a place where teams can build and maintain living work products.
- Codex Cloud lets coding tasks continue from the cloud even when your laptop is closed.
- Agents API + computer use brings OpenAI's multi-agent and computer-control capabilities to developers.
- Plugin extensions let developers build sidebar homes, interactive panels, and file viewers directly inside ChatGPT.
- MCP Events let connected software trigger automations when something happens.
- Sign in with ChatGPT, Pro 500, and the OpenAI Marketplace extend ChatGPT from an app into an ecosystem.
The rest of this guide explains all of it.
---
DevDay is OpenAI's annual event for developers and technical builders.
The 2026 event took place on Tuesday, September 29, 2026 at Fort Mason in San Francisco. OpenAI positioned the event around technical sessions, product demos, APIs, tools, and hands-on discussions with the teams building OpenAI's products.
The opening keynote featured Sam Altman and was livestreamed publicly. The in-person program included developer sessions, workshops, breakout programming, and a closing session.
OpenAI says DevDay 2026 was its largest DevDay so far, with more than 20 major announcements.
That matters because DevDay is not the same thing as a normal ChatGPT release-note update. It is where OpenAI tends to expose the platform direction: the models developers can build on, the primitives OpenAI itself is using, and the surfaces that may become more important over the next year.
Read the official event page here:
Read OpenAI's complete recap here:
---
The easiest way to misunderstand DevDay 2026 is to treat every announcement as a separate feature.
They fit together.
Dots are persistent agents. Space gives those agents and humans shared context. Pages give them shared documents. Team Tasks give them recurring work. MCP Events let external software wake the system up. Plugins give the system tools and interfaces. Codex Cloud lets coding work continue independently of your laptop. The Agents API lets developers build similar patterns into their own products.
So the architectural direction is larger than "ChatGPT got more features."
The direction is:
conversation → persistent context → tools → ongoing agents → event-driven work → collaborative outputs
That is not a guarantee that every traditional app disappears. It is a useful way to understand why OpenAI announced these products together.
Now, here is every major announcement.
---
What Dots are
Dots are OpenAI's new always-on agents.
OpenAI describes them as agents that can get to know what matters to you, keep working on your behalf, and take ongoing responsibilities off your plate.
Each Dot is powered by GPT-6 Astra and gets its own cloud computer and browser. Through OpenAI's plugin ecosystem, Dots can connect to more than 4,000 apps.
A Dot can work across multiple projects instead of forcing you to maintain one chat thread per job.
You can message or call your Dot in ChatGPT. OpenAI also says Dots can be reached through Slack and Microsoft Teams, with texting planned later.
What makes Dots different from a normal ChatGPT conversation
A normal ChatGPT conversation is primarily reactive. You open it, provide a task, and continue the interaction.
Dots are designed for persistence.
OpenAI says Dots can:
- work toward goals 24/7;
- manage several projects at once;
- learn from your feedback and preferences;
- message you with progress or questions;
- perform work on their own cloud computer;
- use connected applications;
- carry context across supported communication channels;
- continue working while you are doing something else.
The important distinction is that a Dot is not merely a saved prompt or recurring timer. It is meant to behave more like an ongoing delegate.
Example uses OpenAI gives
OpenAI's launch page describes examples including:
- noticing recurring product feedback and preparing tested fixes;
- adapting launch materials when the product scope changes;
- rerunning scientific analysis when new data arrives;
- updating a sales proposal as customer requirements evolve;
- turning new interview transcripts into clips, notes, and social drafts;
- detecting that an invoice was forgotten, preparing it, and sending it after approval.
For creators, the content-production example is particularly concrete. OpenAI describes a Dot learning a creator's voice and audience, then processing new interview transcripts into clips, show notes, and social posts for approval.
Controls and safety
OpenAI says each Dot runs on its own cloud computer. Your local computer remains separate unless you explicitly connect it.
Dots can use saved passwords for supported websites without exposing the password itself to the model.
OpenAI also describes a mode it calls proactive research. When you are not actively working with a Dot, it can use already-connected apps in a restricted read-only mode to look for ways to help.
Users can configure rules around actions:
- allow a class of actions;
- require approval;
- block actions.
OpenAI also uses automated action review for actions that could affect accounts or share information.
Certain sensitive actions, such as changing a password, remain user-controlled.
Specialist Dots
OpenAI also previewed specialist Dots for organizations.
These are meant to have their own identity, credentials, hardware, and access rules so a company can assign a Dot a defined organizational responsibility.
OpenAI says it is beginning with focused enterprise pilots.
The company is also working with Microsoft to integrate specialist Dots with Microsoft Agent 365 governance and security controls.
Availability
OpenAI says Dots are rolling out to Pro and Business Premium users in eligible markets.
Enterprise, Edu, and Healthcare users can try the beta when an administrator enables it.
OpenAI says the first Dot is included with eligible plans and that normal conversations with your Dot do not draw down ordinary ChatGPT usage limits. Work that the Dot starts in Codex or ChatGPT Work still counts toward those tools' limits.
Read more
---
What it is
GPT-6.1 Sol is an upgrade to GPT-6 Sol.
OpenAI's positioning is simple: bring much of Astra's intelligence to a lower-cost model that can be used for everyday professional workloads and agents.
OpenAI specifically highlights:
- agentic coding;
- computer use;
- professional work;
- scientific research;
- factuality.
Why OpenAI is emphasizing price-performance
OpenAI says GPT-6.1 Sol approaches GPT-6 Astra on several demanding evaluations while costing substantially less.
Its standard API prices are:
- $2 per million input tokens
- $0.10 per million cached input tokens
- $10 per million output tokens
OpenAI describes that as one-fifth of Astra's standard input and output token prices.
For developers building long-running agents, this matters because repeated tool use and context reuse can turn small per-token differences into large operating-cost differences.
Coding
OpenAI reports GPT-6.1 Sol matching GPT-6 Astra on its cited DeepSWE evaluation at roughly one-fifth of the cost.
It also reports a meaningful gain over GPT-6 Sol.
Treat benchmark claims as benchmark claims: they measure specific test conditions, not every coding task you will encounter. The practical point is that OpenAI is explicitly positioning 6.1 Sol as a strong agentic-coding model below Astra's cost tier.
Professional work
OpenAI highlights PDF understanding and business workflow execution.
It reports GPT-6.1 Sol performing strongly on GDP.pdf, which evaluates difficult professional document questions, and AutomationBench, which evaluates end-to-end business workflows across many tools.
Computer use
GPT-6.1 Sol is also positioned as a major computer-use upgrade over GPT-6 Sol.
This matters for workflows where the model has to operate software rather than simply generate text.
Scientific research
OpenAI reports substantial improvements on Terminal-Bench Science.
The company still says GPT-6 Astra should be used for the most difficult scientific research tasks, but 6.1 Sol is designed to offer much of the useful capability at far lower cost.
Factuality
OpenAI also reports a lower rate of factual errors than GPT-6 Sol on a difficult internal evaluation set based on user-flagged errors.
That does not mean GPT-6.1 Sol is error-free. OpenAI explicitly notes that the evaluation set is intentionally difficult and not representative of typical conversations.
Availability
GPT-6.1 Sol is available in:
- ChatGPT Work;
- Codex;
- the OpenAI API.
OpenAI says it is available to Plus, Pro, Business, Enterprise, and Edu users in Work and Codex.
At launch, it is not yet available in normal Chat.
The API model name is:
gpt-6.1-sol
Read more
Introducing GPT-6.1 Sol — OpenAI
---
What it is
Ultrafast is a premium inference speed tier.
The goal is not to make a model smarter. The goal is to make token generation substantially faster for workloads where waiting is expensive.
OpenAI says Ultrafast can provide:
- up to 8× faster token generation in Codex;
- up to 6× faster token generation in the API.
OpenAI lists GPT-6 Astra Ultrafast as available at launch and says GPT-6.1 Sol Ultrafast is coming soon.
Why this matters
Not every AI task benefits equally from raw output speed.
If you are waiting for a long code-generation loop, interactive agent, real-time tool workflow, or customer-facing response, latency can become a product constraint.
A premium speed tier lets companies trade more money for lower waiting time.
Availability
OpenAI says GPT-6 Astra Ultrafast is available in:
- the API;
- ChatGPT Work;
- Codex.
For ChatGPT, it is tied to Pro 500 and Enterprise access.
Read more
Ultrafast mode — OpenAI API documentation
---
What it is
OpenAI introduced a broader enterprise privacy umbrella called Private Intelligence.
The stated goal is to let organizations use frontier AI while increasing confidence that sensitive data remains protected.
Two parts were highlighted:
- Zero Data Retention with Private Safety Processing
- Private Inference, which OpenAI says is coming in the fall
Private Safety Processing
OpenAI says Private Safety Processing allows automated safety review without OpenAI personnel accessing the underlying customer content.
For organizations with strict data-governance requirements, this is important because safety systems themselves can create data-handling questions.
Private Inference
OpenAI describes Private Inference as combining confidential computing with strict, verifiable controls.
The important word is preview. OpenAI said Private Inference is coming later, so do not treat it as a generally available production feature until the official documentation says it is available.
Read more
Zero Data Retention with Private Safety Processing — OpenAI API
---
What it is
Codex can now run in reusable cloud development environments.
You can start or continue coding tasks from:
- desktop;
- web;
- mobile.
Each task gets an isolated workspace.
The cloud environment can contain your repositories, tools, dependencies, and approved settings so Codex does not have to reconstruct the entire development environment every time.
The important behavior change
Codex tasks can continue even while your computer is asleep.
This shifts Codex from a coding assistant that is tightly attached to your local session toward a coding worker that can operate remotely.
Team use
Reusable environments also matter for teams. A shared environment can reduce the setup drift that happens when every developer or agent runs a slightly different stack.
Workspace administrators still control the relevant cloud-access settings and permissions.
Availability
OpenAI lists Codex Cloud for Plus, Pro, Business, Healthcare, Education, and Enterprise.
Plan limits and workspace policies still apply.
Read more
ChatGPT Business release notes
---
What changed
OpenAI refreshed the Codex command-line interface.
The new CLI includes:
- voice control for starting and steering tasks;
- a new
/agentsview; - easier delegation across multiple tasks;
- improved session resume behavior;
- better prompt editing;
- worktree improvements;
- a cleaner terminal interface.
Why `/agents` matters
The CLI is moving from a single-session interaction model toward multi-agent orchestration.
If you are running multiple coding jobs, the useful problem is no longer "how do I ask Codex to change this file?" It becomes "how do I delegate, monitor, redirect, and reconcile several coding tasks without losing track of them?"
The /agents view is aimed at that coordination problem.
Availability
OpenAI says the refreshed CLI is available on all plans.
Read more
---
What it is
OpenAI introduced a richer code-review experience in the ChatGPT desktop app.
You can:
- read change summaries;
- inspect diffs;
- ask Codex questions about potential issues;
- review work before sharing feedback on GitHub pull requests or GitLab merge requests.
Automatic reviews
Codex can also perform a first-pass review in the cloud while you are away.
This is another example of the DevDay theme: work that used to depend on a synchronous user session can now continue in the background.
Availability
OpenAI says Code Review is available on all plans.
Read more
---
What it is
Codex Security Cloud is a security-focused cloud workflow for defensive software teams.
OpenAI says it can:
- scan entire GitHub repositories;
- run scans on demand;
- run scans on a schedule;
- monitor new commits;
- investigate findings;
- remove duplicates;
- prepare fixes in the cloud.
The work can continue even when the developer's laptop is closed.
Daybreak access
OpenAI says Codex Security Cloud includes access to models offered through Daybreak Blue without requiring a separate Daybreak application.
This is aimed at authorized defensive-security work.
Availability
OpenAI lists Codex Security Cloud for Pro, Business, Enterprise, and Edu users on desktop and web.
Read more
Codex Security Cloud setup — ChatGPT Learn
---
What it is
The Decisions API is a specialized API for real-time classification and routing.
Instead of asking a general-purpose model an open-ended question, the developer defines:
- a specific question;
- a finite set of allowed answers;
- context supplied as text or images.
The system returns one of the developer-defined answers.
OpenAI says the API focuses Luna's intelligence on these bounded decisions.
Example uses
OpenAI gives examples such as:
- classifying content;
- routing requests;
- selecting an agent's next action.
This is useful when your product does not need a paragraph. It needs a reliable choice from a known menu.
For example:
refund_request | bug_report | sales_lead | other
or:
approve | send_for_review | reject
A bounded decision interface can be cheaper and easier to validate than asking a general model to generate unconstrained text.
Availability
OpenAI says Decisions API is in limited preview with broader release planned in the following days.
Because rollout status can change quickly, check the official DevDay recap or API changelog before building around it.
Read more
OpenAI DevDay 2026 Recap — Decisions API section
---
What changed
OpenAI's Agents API now supports computer use.
That means developers can build agents that operate software interfaces as part of a workflow.
The updated Agents API also includes capabilities associated with Codex's agent architecture, including:
- multi-agent behavior;
- tool search;
- tool calling;
- context compaction.
OpenAI runs the underlying infrastructure.
Why this matters
Traditional API integrations work best when every application exposes a clean API.
Real software is messier.
Sometimes a task requires interacting with a web interface, desktop application, legacy system, or workflow that was designed for a human operator.
Computer use lets the agent operate those interfaces when programmatic APIs are unavailable or insufficient.
This is powerful and should be deployed carefully. Computer-use agents can make mistakes, and consequential actions should have approval boundaries.
Availability
OpenAI says the capability is available through the API and in Codex and ChatGPT Work on Pro 500 and Enterprise.
Read more
Computer use — OpenAI Agents API
---
What it is
OpenAI and Amazon are extending OpenAI's agent technology into AWS.
Amazon Bedrock Managed Agents combines OpenAI frontier models and agent infrastructure with AWS-native security and deployment controls.
AWS describes the managed runtime as handling:
- inference;
- memory;
- skills;
- agent execution inside the customer's environment.
Why companies care
Many large organizations already standardize on AWS for identity, logging, networking, and security.
For them, the question is often not "which model is best?" It is "can we deploy this inside the infrastructure and governance model we already operate?"
Bedrock Managed Agents is designed to answer that question for OpenAI-powered agents.
Read more
Amazon Bedrock Managed Agents, powered by OpenAI
---
What they are
OpenAI is opening up more of the platform it uses to build native ChatGPT experiences.
With plugin extensions, developers can give a plugin:
- a home in the ChatGPT sidebar;
- interactive panels that sit alongside the conversation;
- custom viewers for file types the plugin supports.
Why this is bigger than a tool call
Earlier plugin integrations were easy to imagine as invisible functions: ChatGPT sends data to a service, the service returns something, and ChatGPT writes an answer.
Extensions allow the plugin to have an interface.
That means a product can live partly inside ChatGPT instead of forcing the user to leave the conversation for every interaction.
A project-management plugin could expose a board. A design plugin could expose a visual editor. A data tool could expose an interactive dashboard. A specialized file format could have its own viewer.
The point is not that every app should be rebuilt inside ChatGPT. The point is that developers now have a richer native surface.
Availability
OpenAI says plugin extensions are available to all plans.
Read more
Plugin Extensions — OpenAI Developers
---
What changed
OpenAI also changed how plugins are built and discovered.
The DevDay announcement includes:
- Plugin Creator to help developers build plugins;
- a redesigned submission flow;
- clearer submission feedback;
- improved plugin ranking;
- improved recommendations in the directory and in conversations.
Users still choose which plugins to use and approve the access each plugin receives.
Why this matters
A platform is only useful if developers can ship and users can find what they shipped.
Better creation tools lower the supply-side friction.
Better recommendations and discovery lower the demand-side friction.
Taken together with extensions, OpenAI is making plugins look more like a real application ecosystem than a hidden connector layer.
Read more
OpenAI Plugins developer documentation
---
What changed
OpenAI says supported ChatGPT plugins can now be added to Sites.
A Site can therefore become more than a static page. It can expose tools that teammates use with their own connected data and permissions.
OpenAI also says automations are becoming easier to add and manage inside Sites so shared information can stay current.
Team behavior
A key detail is that teammates use the same app with their own permissions.
The Site is shared, but identity and connected-data permissions remain user-specific.
That matters for internal tools, dashboards, operations portals, or team utilities where a common interface should not imply universal access to everybody's data.
Availability
OpenAI lists this for Business, Enterprise, Healthcare, and Edu plans.
Read more
---
What it is
OpenAI announced support for the proposed MCP Events specification.
The core idea is simple:
A connected application can emit an event, and that event can start an automation.
This is important because most AI workflows today begin with the user explicitly asking for something.
MCP Events enables:
something happened → agent wakes up → reads relevant context → does the configured work
Example
OpenAI gives the example of watching for new tasks on a project board.
When a task appears, ChatGPT could:
- receive the event;
- read the linked project documents;
- draft a plan;
- continue the configured work even while you are away.
Other possible patterns include:
- new support ticket → summarize and route;
- new customer lead → research and prepare context;
- new file → extract data and update a dashboard;
- new pull request → run a review;
- new calendar event → prepare a briefing.
The exact actions depend on the plugin, permissions, and automation rules.
Availability
OpenAI says MCP Events support is available to all plans.
Read more
MCP Events — OpenAI Developers
---
What it is
ChatGPT Space is a shared environment for people and AI to work together around ongoing projects.
A Space can contain shared knowledge and work products that teammates, ChatGPT, and Dots build on together.
OpenAI says ChatGPT can also keep the Space organized according to instructions you provide.
Why it is different from a group chat
A group chat is primarily a communication stream.
A Space is closer to a persistent project environment.
The project can include:
- shared context;
- Pages;
- presentations;
- tasks;
- agents;
- connected tools;
- team members.
The goal is to reduce the need to rebuild context every time work changes hands.
Availability
OpenAI says Space is available to Pro, Business, and Enterprise on desktop and web.
On mobile, users can find, read, and share pages. Mobile creation and editing are coming later.
Read more
---
What they are
Pages are a new kind of document designed for collaboration between humans and agents.
OpenAI says anything you can do in ChatGPT can be brought into a Page, including:
- writing;
- research;
- charts;
- images;
- information visualization.
You can create a Page from a conversation and then invite teammates to contribute ideas and feedback.
Why Pages matter
Chats are excellent for exploration.
Documents are better for durable state.
Pages are OpenAI's attempt to connect those two modes so the output of an AI conversation does not have to be copied manually into a separate document before a team can use it.
A Page can become the maintained artifact.
Think:
- project brief;
- research memo;
- living strategy;
- dashboard;
- product requirements;
- content calendar;
- operations document.
Availability
OpenAI lists Pages for Pro, Business, and Enterprise.
Read more
---
What they are
OpenAI announced collaborative slide creation inside ChatGPT.
The feature is designed to let a team and agents work on a deck together.
OpenAI says users will be able to:
- start from a conversation;
- use their own template;
- have multiple people and agents edit at once;
- leave comments;
- present inside ChatGPT;
- export to PowerPoint;
- export to Google Slides with formatting intact.
Availability
This was announced as coming in the weeks after DevDay for Pro, Business, and Enterprise.
Do not assume it is already enabled in your account just because it was announced.
Read more
---
What changed
Business and Enterprise users can now create teams inside ChatGPT and share work across them.
OpenAI says teams can share:
- Pages;
- slides;
- plugins;
- spreadsheets;
- recurring tasks.
Team Tasks
The more important part is recurring work.
A team can delegate recurring responsibilities such as weekly project updates.
ChatGPT can use connected tools to gather information and act:
- on a schedule;
- in response to events;
- after a new email;
- after a Slack message;
- when another supported trigger occurs.
Teammates can refine the shared instructions as priorities change.
Why this matters
Most automation systems fail when nobody owns the automation's instructions after setup.
Team Tasks make the instruction set itself collaborative.
That means the automation can evolve with the team rather than becoming an abandoned black box.
Availability
OpenAI lists team creation and shared tasks for Business and Enterprise.
Read more
Teams and Team Tasks — ChatGPT Learn
---
What it is
ChatGPT can now be invoked directly inside Slack and Microsoft Teams.
Users can mention @ChatGPT in:
- a channel;
- a thread;
- a direct message.
ChatGPT can use tools connected by an administrator or tools connected by the user under that user's permissions.
Shared context
Other teammates can add context and refine the result inside the same workplace conversation.
OpenAI says teammates can participate without each person needing an individual ChatGPT license, subject to the workspace setup.
Why this matters
A major source of AI friction is context switching.
A team discusses something in Slack, then one person copies the thread into ChatGPT, then copies the answer back.
Native access collapses that loop.
Availability
OpenAI lists this for Business and Enterprise.
Read more
ChatGPT in Slack and Microsoft Teams
---
What it is
The new Meetings plugin takes meeting notes and saves personalized summaries with action items into ChatGPT Space.
You can keep summaries private or share them with your team.
Afterward, ChatGPT can use the meeting output to:
- update a project plan;
- draft follow-up communication;
- connect decisions to ongoing work.
Audio handling
OpenAI says meeting audio is deleted after the notes are ready and cannot be replayed or accessed afterward.
That is an important behavior to understand before relying on the tool. The durable output is the notes, not an archived recording.
Availability
OpenAI says the Meetings plugin is in beta in the ChatGPT desktop app on macOS for Pro and Business users, with Enterprise coming later.
Read more
The Meetings plugin in ChatGPT — OpenAI Help Center
---
What they are
Shareable profiles let people showcase the things they have built in the ChatGPT ecosystem.
A profile can surface:
- Sites;
- plugins;
- other reusable creations.
OpenAI also says teammates can discover shared skills inside a workspace.
Skills sharing remains limited to the workspace.
Why this matters
This is a distribution layer.
If plugins and Sites are becoming reusable software, creators need an identity and discovery surface around what they build.
A shareable profile becomes part portfolio, part app shelf, and part reusable-tool directory.
Availability
OpenAI lists profiles for Free, Go, Plus, Pro, Business, and Enterprise.
Read more
Shareable profiles in ChatGPT — OpenAI Help Center
---
What it is
OpenAI is extending ChatGPT identity and plan usage into third-party products.
There are two related ideas:
- Sign in with ChatGPT — use your ChatGPT identity to sign into participating tools.
- Plan usage in participating tools — eligible ChatGPT usage can be applied inside supported partner products.
OpenAI says it is launching plan-usage support across 16 partners, including:
- Cognition's Devin;
- Notion;
- Vercel;
- T3;
- OpenClaw;
- Dactyl.
The exact partner list can change, so use the official Help Center page as the current source.
Why it matters
Today, AI products often require separate:
- accounts;
- billing relationships;
- model-provider keys;
- usage limits.
OpenAI is trying to make your ChatGPT account function more like a portable identity and allowance.
This does not mean every third-party product is free with ChatGPT. Participating products and eligible usage rules control what actually counts.
Availability
OpenAI says identity is available globally.
Plan usage is available to Plus and Pro users in participating tools.
Read more
Using your ChatGPT plan in other apps and sites — OpenAI Help Center
---
What it is
OpenAI introduced a new Pro 500 subscription tier.
The name refers to the $500-per-month tier.
OpenAI says it provides:
- its highest usage allowance;
- 25× the ChatGPT Plus allowance;
- access to Ultrafast.
This tier is aimed at people whose AI usage is closer to heavy professional infrastructure than occasional chat.
Who it is for
The practical question is not "is Pro 500 the best plan?"
The practical question is whether your workload is constrained by:
- usage ceilings;
- agent throughput;
- Codex throughput;
- model speed.
If not, paying for more capacity does not create value by itself.
Availability
OpenAI says Pro 500 is available now.
Read more
About ChatGPT Pro tiers — OpenAI Help Center
---
What it is
OpenAI Marketplace is an enterprise purchasing and partner-software program.
Eligible enterprise customers can apply part of an existing OpenAI commitment toward approved partner products.
OpenAI's initial group includes 32 partners.
The company names partners across categories including:
Creative - Adobe - Figma
Customer experience and enterprise software - Sierra - Decagon - HubSpot - Salesforce - ServiceNow
Legal - Harvey - Legora
Cybersecurity - Palo Alto Networks - CrowdStrike
Infrastructure / open-source model serving - Baseten
Why this matters
Large enterprise AI budgets do not live in isolation.
If a company already has a major OpenAI commercial commitment, Marketplace gives it more flexibility to direct part of that spend toward approved software that sits around the OpenAI ecosystem.
For OpenAI, this also deepens the ecosystem around its platform.
Availability
OpenAI says eligible enterprise customers can express interest.
Read more
---
The individual products matter, but the composition matters more.
Here are the major shifts.
Shift 1: From sessions to persistent agents
ChatGPT began as something you opened when you had a question.
Dots are explicitly designed to remain active and keep context across time.
That changes the interaction model from:
ask → answer → leave
to:
delegate → monitor → review → redirect
Shift 2: From individual chat to shared work state
Space, Pages, slides, Teams, and Team Tasks create persistent shared artifacts.
Instead of AI living in a private side conversation, the AI can participate in the same work environment as the team.
Shift 3: From scheduled automation to event-driven automation
Scheduled tasks are useful when you know the clock time.
MCP Events matters because real work often begins when something happens, not at 9:00 AM.
Examples:
- customer replied;
- issue opened;
- form submitted;
- task created;
- document changed;
- new email arrived.
This makes automations more responsive to real-world state.
Shift 4: From text plugins to native mini-apps
Plugin extensions add interfaces.
That expands what can happen inside ChatGPT because not every useful interaction should be expressed as prose.
A map is better as a map.
A board is better as a board.
A chart is better as an interactive chart.
A specialized file is better in a viewer.
Shift 5: From local coding agent to cloud development worker
Codex Cloud makes coding tasks less dependent on a single laptop session.
Combine that with multi-agent CLI controls, cloud review, and security scanning and Codex increasingly looks like a persistent software-work environment.
Shift 6: From one OpenAI app to an ecosystem
Sign in with ChatGPT, plan portability, plugins, profiles, Sites, Marketplace, AWS agents, Slack, Teams, and developer APIs all push in the same direction:
OpenAI wants ChatGPT and its models to be a platform layer that appears across other software, not a destination you always have to visit manually.
---
This is a simplified launch-day snapshot. Availability can change after publication, so click the linked source before making a purchase or promising a feature to a client.
| Announcement | Launch status | |---|---| | Dots | Rolling out to Pro and Business Premium; enterprise beta can be admin-enabled | | GPT-6.1 Sol | Available in Work, Codex, and API; not yet normal Chat at launch | | GPT-6 Astra Ultrafast | Available in API and Work/Codex on Pro 500 and Enterprise | | GPT-6.1 Sol Ultrafast | Coming soon | | Private Safety Processing | Available documentation; enterprise-focused | | Private Inference | Preview / coming in fall | | Codex Cloud | Available on supported ChatGPT plans | | Refreshed Codex CLI | Available | | Code Review | Available | | Codex Security Cloud | Available to supported paid/workspace plans | | Decisions API | Limited preview | | Agents API computer use | Available through supported API / premium product surfaces | | Bedrock Managed Agents | AWS offering; check AWS status | | Plugin extensions | Available | | Plugin Creator / discovery improvements | Available / rolling through plugin ecosystem | | Plugins in Sites | Business and enterprise-oriented plans | | MCP Events | Announced available across plans, subject to supported plugin | | ChatGPT Space | Pro, Business, Enterprise on desktop/web | | Pages | Pro, Business, Enterprise | | Collaborative slides | Coming in following weeks | | Teams and Team Tasks | Business and Enterprise | | @ChatGPT in Slack/Teams | Business and Enterprise | | Meetings plugin | Beta for Pro and Business on macOS desktop; Enterprise later | | Shareable profiles | Free through Enterprise | | Sign in with ChatGPT | Identity global; plan usage through participating partners for Plus/Pro | | Pro 500 | Available | | OpenAI Marketplace | Enterprise interest / eligibility |
Primary source for this table:
---
If you are not a developer, do not try to learn every API announcement.
Start with the product changes that alter your daily workflow.
First: Dots
If Dots are available in your account, give your Dot one ongoing responsibility instead of twenty random errands.
Good first responsibilities:
- track an active project;
- prepare a recurring briefing;
- process new creator interviews into content assets;
- watch an inbox or workflow for a defined class of updates;
- maintain research around a single decision.
A narrow responsibility gives you a much better signal about whether the agent is reliable.
Second: Space
Create one Space for a real project.
Put the project's shared material there.
Then use Pages for the actual maintained output instead of repeatedly asking ChatGPT to reconstruct the project from old conversations.
Third: Codex Cloud
If you build software, use one non-critical repository and test the complete loop:
- prepare the environment;
- delegate one bounded task;
- close your laptop;
- return later;
- inspect the diff;
- ask Codex to explain the implementation;
- run your normal verification before merging.
Fourth: event-driven automations
If a plugin you use supports MCP Events, build one automation around a real trigger.
Start with a low-risk read-and-draft workflow before allowing actions.
Example:
new support ticket → read context → draft response → wait for approval
That gives you the benefit of proactive AI without immediately granting broad autonomous control.
---
Developers should separate the announcements into four layers.
Model layer
- GPT-6.1 Sol
- Ultrafast
These change the intelligence/cost/latency choices.
Agent layer
- Agents API with computer use
- Decisions API
- Bedrock Managed Agents
These change how developers orchestrate work.
ChatGPT platform layer
- plugin extensions
- plugin creation and discovery
- Sites hosting plugins
- MCP Events
- shareable profiles
These change how developers distribute software inside ChatGPT.
Collaboration layer
- Space
- Pages
- slides
- team tasks
- Slack/Teams
- Meetings
These create new surfaces where third-party tools and agents may become useful.
The biggest strategic question is not which announcement is coolest.
It is:
Where should your product live if users increasingly expect AI to carry context, invoke tools, and maintain work across surfaces?
That is the question DevDay 2026 puts in front of software builders.
---
A launch-day guide can become misinformation very quickly if it ignores rollout details.
Keep these caveats in mind.
Announced does not always mean enabled in your account
Several features are rolling out gradually, restricted to specific plans, or explicitly coming later.
Always check the current product UI and official documentation.
Dots are autonomous, not infallible
OpenAI explicitly says Dots can make mistakes.
Consequential work should be reviewed.
Autonomy does not remove the need for permissions, boundaries, and human judgment.
Benchmarks are not your workload
GPT-6.1 Sol's benchmark results are useful signals.
They are not a promise that it will match Astra on your exact task.
Run your own evaluation if model choice affects cost or quality materially.
Computer use should not be treated like a deterministic API
A computer-use agent operates interfaces designed for humans.
Interfaces change. Visual recognition can fail. A button can move. A website can produce an unexpected state.
Use approval gates for consequential actions.
Plugin permissions still matter
Native interfaces do not erase security boundaries.
Users still approve plugin access, and workspace administrators can control eligible connections.
Team context is not universal access
Shared Spaces and Sites do not mean every teammate automatically has unrestricted access to every connected account.
Identity and permissions still matter.
"Sign in with ChatGPT" does not mean every product is included in your subscription
Only participating partners and eligible usage count toward plan allowances.
Always read the participating product's terms.
---
If you want to dig deeper without searching, use this list.
Event overview
Agents and models
Codex
Agents API and AWS
- Computer use in the Agents API
- Amazon Bedrock Managed Agents powered by OpenAI
- OpenAI on Amazon Bedrock
Plugins
Collaboration
Identity, plans, and ecosystem
- Shareable profiles
- Using your ChatGPT plan in other apps and sites
- About ChatGPT Pro tiers
- OpenAI Marketplace
---
DevDay 2026 was OpenAI's clearest attempt yet to move ChatGPT from a chatbot you open into a persistent environment where models, agents, people, tools, documents, software interfaces, and external events can all participate in the same ongoing work.
That is the useful mental model.
Do not memorize every feature.
Watch the direction:
AI that waits for you → AI that stays with the work.
Want a hand applying this?
Install the AGI Edge MCP in Codex, then use this as your starting prompt:
“I am reading The AGI Guide DevDay 2026 guide. Ask what kind of work I do, then identify the three DevDay announcements most relevant to me, explain exactly why, and give me one practical experiment I can run with each this week.”
Install the MCP in Codex