OpenAI DevDay 2026: 25 Updates – dots, GPT-6.1 Sol, Ultrafast, Codex & Pro Pricing

On September 29, 2026 (US time), OpenAI held DevDay in San Francisco, announcing 25 modules covering new models, agents, developer tools, plugins, team collaboration, and subscription changes. Some features are already available; others remain in preview or pending rollout.

01 dots: Persistent AI Agents with Cloud Environments

dots are resident agents powered by GPT-6 Astra. Each dot gets its own cloud computer and browser, able to operate across 4,000+ connected applications. They can handle one-off tasks (e.g., restaurant reservations) or continuously drive a project in the background, learning user preferences and acting proactively 24/7. The author created a dot named “石臻Dot” and noted it can research AI trends, verify sources, organize scripts, and advance projects after the user leaves the chat.

Autonomy scope is user-defined: which actions run automatically, which require confirmation, and which are forbidden. Connected apps are also chosen by the user. By default dots use their own cloud computer; local computer connection is optional.

Important limitation: When no one is actively interacting, “proactive research” is limited to read-only tools within connected apps — it cannot send messages, modify app content, or control the computer. Execution still requires task authorization and automatic review.

Availability: First rollout to eligible Pro and Business Premium users; Enterprise, Edu, and Healthcare admins can enable beta (default off). Creation via desktop app or web browser; accessible on web, mobile, and desktop. The first dot for Pro/Business Premium is included at no extra charge, and conversations with it do not count against ChatGPT usage. However, Work or Codex tasks launched by the dot still consume their respective quotas. A temporary quota boost for deeper work applies for the first month.

Enterprise pilot: Specialist dots with dedicated identities and system permissions for tasks like procurement and invoice processing; planned integration with Microsoft Agent 365 governance. This is a limited pilot, not generally available.

Communication channels: ChatGPT, Slack, Teams, and voice; SMS planned for later.

Dot vs. ChatGPT, Work, and Codex

The author provides a decision framework based on official positioning:

ChatGPT Chat: Daily Q&A, search, idea discussion, quick back-and-forth.

ChatGPT Work: Multi-step research and analysis, delivering documents, spreadsheets, presentations, reports, or websites.

Codex: Writing code, debugging, running commands and tests, reviewing changes, handling software projects.

Dot: Continuously learns your goals and preferences, follows up across tasks; can launch or manage Work and Codex tasks when needed.

Example workflow for a content creator: Dot tracks topics and projects, Work assembles research and slide decks, Codex builds collection tools or product features. Work already supports long tasks, scheduled or event-triggered runs; Codex Cloud continues after local sleep. Dot’s distinction is persistent learning, coordination, and follow-up around a single user’s objectives. Capabilities overlap and can be combined.

Quota details from email: Dot conversations and its direct work are included in Pro subscription, not counted against ChatGPT limits. But Work/Codex tasks started by Dot still count normally — not everything via Dot becomes free/unlimited.

02 GPT-6.1 Sol: Near-Astra Performance at One-Fifth Standard Price

GPT-6.1 Sol upgrades GPT-6 Sol, focusing on agentic coding, computer operation, and complex professional tasks. Positioned as delivering near-Astra capability at lower cost.

Price comparison (per million tokens):

Input: GPT-6.1 Sol $2 vs. GPT-6 Astra standard $10

Cached input: GPT-6.1 Sol $0.10 vs. Astra $1 (95% cheaper than Sol’s own regular input, 50% cheaper than GPT-6 Sol cached input)

Output: GPT-6.1 Sol $10 vs. Astra $50

Cached input benefits apps that reuse long contexts, tool definitions, or project materials.

Benchmark highlights (official tests):

DeepSWE v1.1 (real codebase tasks): Sol 6.1 matches Astra at ~1/5 cost.

OSWorld 2.0 offline set (computer operation): Sol 6.1 improves 7 percentage points over GPT-6 Sol, still 2.1 points behind Astra.

AutomationBench medium (multi-step business processes): +4.8 points over GPT-6 Sol.

Scientific workflows: Highest reasoning setting scores over 2× GPT-6 Sol, but Astra still leads on hardest research tasks.

Factual accuracy & intent adherence: On a curated hard set, low-reasoning factual error rate dropped from 11.4% to 7.7% (this set is designed to induce errors, not representative of everyday error rate).

Access: Plus, Pro, Business, Enterprise, Edu users in ChatGPT Work and Codex; developers via API with model ID gpt-6.1-sol. Standard Chat does not yet have this model.

03 Ultrafast: Dedicated High-Speed Tier

Ultrafast is a premium speed tier. Official ceiling: Codex token generation up to 8× faster (300 tokens/sec), API up to 6× faster. This is token generation speed; end-to-end task time may not improve 8× because browser loads, test runs, and external tool calls still take time. Value lies in reducing wait time for long tasks that repeatedly call the model.

Availability: GPT-6 Astra Ultrafast live in API, ChatGPT Work, Codex; GPT-6.1 Sol Ultrafast coming soon. Personal subscriptions require Pro 500; eligible Enterprise/Edu workspaces also allowed (permissions and workspace credits apply). Plus, Pro 100, Pro 200, and Business cannot unlock via credit purchase at launch.

Cost: Astra Ultrafast consumes subscription quota at 8× standard rate; when using credits, charged at 6× standard rate. These are billing multipliers, not speed multipliers.

04 Private Intelligence: Extra Enterprise Data Protection

Combines zero-data-retention with Private Safety Processing (automated safety review without OpenAI staff accessing raw content) and Private Inference (confidential computing with verifiable controls to protect data during inference). Targeted at enterprises handling sensitive internal data. Private Inference preview slated for autumn; access requires contacting OpenAI — not a default feature for personal accounts.

05 Cloud Codex: Cross-Device Continuity

Codex runs locally, remotely from phone, or in the cloud from multiple devices. Reusable dev environments let teams share pre-configured settings and permissions, cutting setup time per task. Available to Plus, Pro, Business, Healthcare, Education, Enterprise. Tasks run in cloud; users can monitor and steer from other devices after leaving their computer.

06 Codex CLI: Voice Commands in Terminal

CLI adds voice launch and guidance. New /agents view assigns work and tracks multiple agents. Improvements to prompt editing, session recovery, worktree usage, and terminal readability for long sessions. Available on all plans; CLI users don’t need to switch to GUI for these interactions.

07 Code Review: Centralized Diff Review with Codex Pre-Check

New code review experience in ChatGPT desktop app: view change summaries, browse diffs, ask Codex for potential issues. Supports GitHub Pull Requests and GitLab Merge Requests. Automated review runs in cloud first, then human review and feedback. All plans. Focus is change review; next feature addresses continuous vulnerability scanning.

08 Codex Security Cloud: Continuous Vulnerability Scanning

Scans entire GitHub repos on demand or on schedule, continuously checks new commits, investigates findings, deduplicates, and prepares fix proposals. Runs entirely in cloud — local machine can be off. Access via Codex desktop and web plugins for Pro, Business, Enterprise, Edu. Includes Daybreak Blue model access (no separate Daybreak application needed). Does not grant Daybreak Red capabilities. Generated patches are proposed fixes requiring human verification for correctness and regression risk.

09 Decisions API: Fast Classification Among Defined Options

Developers define a question and a closed set of candidate answers; Luna judges based on provided text or image context. Suited for content classification, request routing, or deciding an agent’s next path (e.g., categorize incoming request as “pre-sales”, “support”, “needs human”). Limited preview at launch, broader rollout planned within days.

10 Agents API: Managed Agents with Computer Use

Agents API adds computer use. Developers let agents operate software to complete tasks, leveraging Codex’s multi-agent, tool search, tool calling, and context compression. Application defines task, tools, data, and progress display; OpenAI manages underlying execution infrastructure. Available via API; computer-use capabilities also in Codex and ChatGPT Work for Pro 500 and eligible Enterprise users.

11 Bedrock Managed Agents: OpenAI Agents on AWS

Partnership with Amazon brings Agents API core into Bedrock Managed Agents, adding AWS-native customization and resource integration. Teams already on AWS can deploy these agents internally, connecting existing resources. Exact enablement and scope per AWS product documentation.

12 Plugin Extensions: Sidebar Panels and File Viewers

Plugins can now have sidebar entry points, interactive panels, or viewers for specific file formats. Users operate panel while chatting; developers can embed richer app experiences inside ChatGPT. All plans.

13 Plugin Creation & Discovery: Smoother Build, Submit, Find

Plugin Creator assists building; redesigned submission flow gives clearer feedback. Directory and in-chat ranking/recommendations improved for discoverability. Users still approve which plugin and what access. All plans.

14 Sites + Plugins: Per-User Data in Shared Apps

Sites-created apps can now connect supported ChatGPT plugins. Coworkers accessing the same app use their own connected data and permissions. Automated addition/management keeps shared info current. Business, Enterprise, Healthcare, Edu.

15 MCP Events: Event-Driven Automation

Support for proposed MCP Events spec. When connected apps emit events, plugins can trigger automated tasks. Example: new task on project board → ChatGPT reads linked docs, drafts plan automatically. All plans; actual triggers depend on plugin support and authorization. Spec is still a proposal, not a finalized standard.

16 ChatGPT Space: Shared Team Workspace

Space puts team members, ChatGPT, and dots in one collaboration space built on shared knowledge. Users can set organization rules for ChatGPT to maintain space structure for easier retrieval and continuation. Pro, Business, Enterprise on web and desktop; mobile supports search, read, share — create/edit coming later.

17 Pages: Collaborative Documents from Chat

Pages are new documents designed for human-agent co-work: write content, research, generate charts/images, embed visualizations. Create from chat, invite colleagues to edit/comment, then have ChatGPT or dot continue editing. Better for reports and project plans that need repeated multi-person updates than scattered chat history. Pro, Business, Enterprise.

18 Collaborative Slides: Co-Edit Presentations with AI

Create slides from chat or templates; multiple colleagues and agents edit and comment simultaneously. Present in ChatGPT or export to PowerPoint/Google Slides with formatting preserved. Pro, Business, Enterprise — but English official list marks “coming in the next few weeks”, not day-one.

19 Team Creation & Shared Tasks: Recurring Work like Weekly Reports

Colleagues form teams to share pages, slides, plugins, spreadsheets, and set up recurring tasks. Example: weekly project update — ChatGPT collects info via connected tools, runs on schedule or events (new email, Slack message), team jointly refines instructions. Business and Enterprise.

20 @ChatGPT in Slack & Teams: In-Context Collaboration

Mention @ChatGPT in channels, threads, or DMs to involve it in team work, using admin-connected tools or user-authorized personal tools. Coworkers add context and adjust results in same thread; not every participant needs a personal ChatGPT license. Business and Enterprise; tool access still governed by permissions.

21 Meetings Plugin: Notes to Actions

Records meetings, generates personalized summaries with action items, saves to Space. Summaries can be private or shared; ChatGPT can then update plans or draft follow-ups. Launch beta on macOS desktop for Pro and Business; Enterprise soon. Audio deleted after summary — cannot be replayed or retrieved, so users needing original recordings must plan accordingly.

22 Shareable Profile Page: Showcase Sites & Plugins

Profile page aggregates your created Sites and plugins for discovery, sharing, reuse. Teams can discover shared skills (skill sharing limited to workspace). English list shows Free, Go, Plus, Pro, Business, Enterprise; Enterprise rollout timing varies by language — actual account entrance is authoritative.

23 Sign in with ChatGPT: Identity & Subscription Access for Partner Tools

Log into partner products with ChatGPT account, reducing password management. Identity login globally available; specific availability subject to org settings. Plus/Pro users can use subscription quota in participating partner tools. Initial 16 partners include Devin, Notion, Vercel, T3, OpenClaw, Dactyl; per-partner quota controls provided. Usage counts against your plan limits — review authorization scope and quota allocation before enabling.

24 Pro 500, Pro 200 & Quota Rules

New Pro 500 at $500/month , quota = 25× Plus, includes Astra Ultrafast.

Critical Pro 200 change (email notice): Price stays $200/month, but from October 30, 2026 quotas shift:

ChatGPT Work & Codex included quota: from 20× Plus down to 10× Plus.

Chat GPT-6 Pro message cap: from 200/week down to 100/week.

Email cites GPT-6.1 Sol’s efficiency gains as justification — fewer round trips, handles previously impossible work. For heavy Work/Codex users, price unchanged but included quota halved is a significant change.

One-time grant of 62,500 usage points (valued $2,500) , expiring Dec 31, 2026, viewable in Settings → Usage, subject to service quota terms. Not a cash refund; not guaranteed for every Pro user. Author had not yet received the grant at time of writing.

Pro 200 now open to new subscribers; new subs without legacy quota transition get adjusted quotas; retaining old quota does not auto-grant Ultrafast. Pro 500 provides 25× Plus quota and Ultrafast (up to 8× token speed). OpenAI commits to no return of 5-hour rolling limits, allowing concentrated weekly quota use. Weekly quota still exists — not unlimited. Email confirms Pro subscription now includes Dot.

25 OpenAI Marketplace: Enterprise Procurement via Committed Spend

Eligible customers can allocate a portion of existing OpenAI committed spend to purchase approved partner software. First 32 partners span creative, customer service, legal, cybersecurity (Figma, Adobe, Salesforce, Harvey, CrowdStrike, etc.). Involves enterprise procurement and contract commitments; personal Pro subscriptions cannot directly use this.

Practical guidance: If you use Codex or Work daily, evaluate GPT-6.1 Sol first; for long-running work tracking, consider dots; if you frequently wait on model output, then assess Ultrafast. Team collaboration users should prioritize Space, Pages, and event automation. Confirm your plan and access points before deciding to upgrade.

Early user jaatster reports dot understands work context; developer Rudrank Riyam plans to test GPT-6.1 Sol. Both are anecdotal initial impressions, not consensus. Updates are fresh; official demos and tests cannot replace long-term real-world validation. Key open questions: Can dots reliably sustain multi-day tasks? How much does Ultrafast actually reduce end-to-end work time? More independent use cases needed.