OpenAI DevDay 2026: Dots, GPT-6.1 Sol, Codex and Microsoft 365 Admin Impacts

This rundown covers the whole keynote. It separates what OpenAI's own help documentation confirms from what comes only from keynote coverage, and it fixes one date error that appears in some recaps.

Dots: always-on agents with their own cloud computers​

The headline launch is dots, persistent agents powered by GPT-6 Astra. According to RuntimeWire's keynote rundown, each dot gets its own cloud computer, can work across connected apps, keeps context between conversations and can juggle several ongoing projects. You can reach a dot through ChatGPT, Slack or Microsoft Teams, including voice calls. Texting is coming later. With your permission, a dot can also connect to your laptop.

Other outlets back up the main points. The Next Web reports that dots run on GPT-6 Astra and connect to over 4,000 apps. It also reports that Pro users in the European Economic Area, Switzerland and the UK will not get dots for now. SQ Magazine adds that Pro subscribers must be over 18, that Business Premium is rolling out worldwide, and Enterprise admins have to switch dots on.

Usage and safeguards:

• RuntimeWire says your first dot comes with no extra subscription charge. Work and Codex tasks it starts still use your normal plan allowance.
• tbreak reports that conversations with a Dot do not count towards ChatGPT usage limits, though tasks it starts in Codex or ChatGPT Work do.
• Background "proactive research" uses connected tools in read-only mode. Any action goes through permissions and approval rules.
• tbreak also reports that OpenAI provides Custom Rules that can allow, block or require approval for particular actions. An auto-review step checks each action against those rules.

Section summary: Dots are real and rolling out now, but availability is uneven. Pro users in the EEA, UK and Switzerland are excluded for now, and managed workspaces need an admin to turn them on.

Specialist dots and Microsoft Agent 365​

This is the part Microsoft admins should watch. OpenAI previewed specialist dots, which are agents for defined business roles with their own identities, credentials and access to company systems. The Next Web says OpenAI tested them internally across procurement, invoice processing, email marketing, customer support and commercial contracting. It also says the company is starting with enterprise pilots and is working with Microsoft to bring specialist dots to Agent 365.

Put plainly, a specialist dot is a non-human account that can act on its own. That brings the usual questions for any account in an Entra-backed tenant:

• How is it provisioned?
• How much access does it get?
• What gets logged?
• Who can shut it off?

SQ Magazine asks the obvious follow-up: what audit trail will Agent 365 give enterprises? So far, neither OpenAI nor Microsoft has published documentation showing how specialist dots will appear in Entra, Purview or the Teams admin tools. Treat the Agent 365 integration as announced, not shipped.

GPT-6.1 Sol, Astra Ultrafast and the new Pro tiers​

GPT-6.1 Sol upgrades GPT-6 Sol. OpenAI says it approaches Astra's performance on coding, computer use and professional work at one-fifth of Astra's standard token prices. RuntimeWire reports these API prices:

The "95% discount" mentioned on stage refers to cached input compared with standard input. That makes reusing context much cheaper than sending fresh input. The API model name is gpt-6.1-sol.

OpenAI's help center confirms that GPT-6.1 Sol is a ChatGPT Work and Codex model and is not available in regular ChatGPT conversations. OpenAI's Business and Enterprise release notes describe a staged rollout that starts with Pro users and expands to Plus, Business, Enterprise and Edu. Enterprise and Edu administrators must enable access. If the model is missing from your picker today, you probably just haven't reached your place in the rollout yet.

Astra Ultrafast is a premium low-latency tier for GPT-6 Astra:

• The keynote cited 300 tokens per second at six times the standard price.
• OpenAI's recap claims up to eight times faster token generation in Codex and up to six times faster in the API.
• Those numbers measure how fast tokens are generated, not how fast a whole task finishes. Tool calls and other steps still take their own time.

OpenAI's usage documentation defines who can use it:

• Pro $500: can use included Work and Codex usage and credits.
• Eligible Enterprise and Edu workspaces: Ultrafast draws on workspace credits and requires the right permissions.
• Plus, Pro $100, Pro $200 and Business: not included at launch, and buying credits on these plans does not enable it.

GPT-6.1 Sol Ultrafast is described as coming soon.

Pricing changes: The new Pro 500 plan costs $500 a month, includes Ultrafast, and offers an allowance OpenAI describes as 25 times Plus usage. RuntimeWire reports that Pro 200 has reopened to new subscribers with a smaller allowance, and that eligible existing subscribers keep their earlier allowance until October 29, 2026. SQ Magazine also calls this a usage cut for Pro 200. I could not find the exact allowance numbers in OpenAI's help pages. If you're on Pro 200, check Settings → Usage before October 29.

Section summary: Sol is the cost-efficient model for most work, Ultrafast is a paid speed option, and Pro 200 subscribers should expect a smaller allowance after October 29.

Developer platform: Decisions, Agents, computer use and AWS​

• Decisions API (limited preview): Uses the Luna model to pick answers from a predefined set, based on text or image input. It's meant for classification, request routing and choosing an agent's next action. On stage, OpenAI described sub-second responses and vision-driven hardware reactions. Broader access is planned "in the coming days."
• Agents API (public beta), with a date correction: OpenAI actually announced it on September 10, 2026, in its own post, so DevDay restated it rather than introducing it. It gives developers the Codex harness as a managed service. OpenAI handles sessions, orchestration, context compaction and recovery. Developers run agents in an OpenAI-managed sandbox, on their own infrastructure, or with sandbox partners such as Cloudflare, Vercel, Modal and Oracle. OpenAI says there is no extra fee for the API itself: you pay for the tokens and tools your agents use.
• Computer use: Agents can navigate websites and operate web applications through an OpenAI-hosted browser, and developers can follow the session's events as it runs. It is also listed in Work and Codex for Pro 500 and Enterprise. An OpenAI-hosted browser is not the same as control of your PC.
• Bedrock Managed Agents: OpenAI and Amazon are combining OpenAI's harness with AWS infrastructure. Amazon says agents run in the customer's environment, inference runs on Bedrock and data stays within AWS. The service handles inference, memory and skills, and it works with Bedrock AgentCore.

Codex: cloud environments, CLI, code review and security​

• Codex Cloud reusable environments: An environment bundles a project's repositories, dependencies, tools and access settings. Each task runs in its own isolated workspace. OpenAI's Enterprise release notes confirm that authorized members can share these environments and that tasks keep running while your computer sleeps. Enterprise cloud-access settings still apply. You have to create and publish an environment on desktop or web before you can pick it for a task on mobile.
• Codex CLI: Adds voice control for starting and steering tasks, a new /agents view for tracking delegated work, and improvements to prompt editing, session resumption, Git worktrees and the terminal interface. OpenAI's help center says Astra requires Codex CLI 0.153.0 or newer.
• Code Review in the ChatGPT desktop app: Pulls PR summaries, changed files, comments and checks into one view. You can inspect diffs and ask Codex to investigate an issue before posting feedback, and automatic cloud reviews can run while you're away. GitHub support is generally available. GitLab merge request support is in preview.
• Codex Security Cloud: Scans connected GitHub repositories on demand or on a schedule and watches new commits. OpenAI says it investigates findings, removes duplicates and prepares fixes in the cloud. It also includes access to models offered through Daybreak Blue without a separate Daybreak application.

Windows note: OpenAI's help center says the new ChatGPT desktop app is available on both macOS and Windows. Voice works in Work and Codex on both platforms. If Astra doesn't show up, OpenAI's fix is:

1. Open the app and select Menu → Check for Updates.
2. Install the update, then fully quit and reopen the app.
3. Check for updates again, and repeat if another update appears.
4. Look at the model picker and Settings → Usage in the account or workspace you plan to use.

Private Intelligence​

For customers handling sensitive data, OpenAI announced Zero Data Retention with Private Safety Processing. Automated safety reviews run without human access to the protected content. Encrypted safety records are kept in customer-controlled storage and reviewed in a hardware-attested runtime.

A separate preview, Private Inference, combines confidential computing with verifiable controls and is due this fall. These privacy claims are OpenAI's own and haven't been independently audited, so compliance teams should ask for the attestation details before relying on them.

Plugins, Sites and MCP Events​

• Plugin extensions: Developers can add sidebar destinations, interactive panels and file viewers to ChatGPT. The keynote demos included Figma designs and comments beside a chat, Photoshop functions inside ChatGPT, and OpenAI's Meetings app. Web extensions for Free and Go users are coming soon, and composer mentions currently require the desktop app.
• Plugin Creator and submission changes: You can build a reusable plugin by describing the workflow in conversation. Developers can now track reviews, see what needs fixing, request human review and update tools without restarting a submission. ChatGPT can also recommend plugins inside conversations.
• Sites with connected data: OpenAI's Business and Enterprise release notes say teammates approve read-only access to their own connected accounts. Admins must turn on Allow use in Sites for each plugin, and that setting is off by default.
• MCP Events: Support for the proposed MCP Events specification lets connected apps notify ChatGPT when something changes, which can trigger an automation. It uses webhooks and needs developer implementation. Connecting an app does not create an automation by itself. For event-triggered Work tasks, Enterprise, Edu and Healthcare admins must first enable Allow event-triggered scheduled tasks, which is also off by default.

Collaboration: Space, Pages, slides, Teams and Meetings​

• ChatGPT Space brings teammates, ChatGPT and dots together around shared knowledge. It is available on Pro, Business and Enterprise, on desktop and web.
• Pages is a new document format for writing, research, charts and interactive work, with comments and real-time editing.
• Collaborative slides is in preview. You'll be able to present in ChatGPT or export to PowerPoint and Google Slides, with availability on Pro, Business and Enterprise planned for the coming weeks. OpenAI's Space product page lists collaborative spreadsheets as coming soon.
• Teams and Team Tasks let coworkers share recurring work that runs on a schedule or in response to events such as an email or Slack message. OpenAI's release notes stress that team membership does not share personal chats or app connections and does not override organizational access controls.
• @ChatGPT in Slack and Microsoft Teams: You can mention ChatGPT in a channel, thread or direct message. RuntimeWire reports that colleagues can take part without each needing a ChatGPT license. OpenAI's docs don't confirm that yet.
• Meetings (beta): Captures online or in-person meetings, saves summaries and action items into Space, and deletes the audio once notes are ready. It launches only in the macOS desktop app for Pro and Business. Enterprise is in limited testing, and Windows, iOS and Android support is planned.
• Shareable profiles display your Sites and activity. They are rolling out gradually, with Enterprise support coming soon.

Sign in with ChatGPT and OpenAI Marketplace​

Sign in with ChatGPT lets Plus and Pro customers use their Work and Codex allowance in participating apps, with limits set per app. RuntimeWire counts 16 tools, including Devin, Notion, Vercel and T3. OpenAI's help page lists a shorter initial group: Airtable, GitLab, HubSpot, Notion, Supabase and Vercel. The difference probably reflects the rollout, so check each app.

The help page also settles the privacy question:

• Identity sign-in shares only your name, email address and profile picture.
• It does not share your conversations, memory, files, tokens or billing information.
• Using your plan allowance in a partner app is a separate permission you can decline.
• Global admins manage these settings in the OpenAI Admin Console under External access.

OpenAI Marketplace (beta) lets eligible enterprises apply part of their OpenAI spending commitment to approved partner software. RuntimeWire names 32 launch partners, including Adobe, Salesforce, ServiceNow, CrowdStrike and Palo Alto Networks. OpenAI's Enterprise release notes add details that change how this works in practice:

• Eligibility is decided per customer, product and purchase.
• Customers contract with and are invoiced by the partner.
• There is no self-service checkout.

The research intern claim​

OpenAI repeated its claim that it has reached its goal of an automated research intern. The milestone first appeared in its September 6 research report. It also said its models can handle more than a third of day-long research tasks. These are OpenAI's claims and haven't been independently verified.

Bottom line​

• Dots: rolling out now to Pro and Business Premium, but Pro users in the EEA, UK and Switzerland are excluded for now, and managed workspaces need an admin to enable them.
• Specialist dots in Agent 365: announced, not shipped. Don't plan around Entra or Purview controls until they're documented.
• GPT-6.1 Sol: in Work, Codex and the API, not in regular Chat, and rolling out in stages.
• Ultrafast: Pro 500 and eligible Enterprise/Edu only. Buying credits on other plans won't enable it.
• Agents API: announced September 10 and restated at DevDay.
• Defaults: Several new data-access and automation settings are off by default, including Allow use in Sites, event-triggered tasks, and Sites/Ads delegated access. Review each one before approving it.
• Meetings: macOS only for now. Windows is planned.