On September 29, 2026, in San Francisco, despite being suffocated by old rival Anthropic in model competition, OpenAI DevDay 2026 still took place as scheduled.
But at this launch event, the new flagship model was still absent.
The GPT-6.1 Astra, originally scheduled for an October release, was urgently pulled by the company twenty-four hours before the opening. The new protagonist stepping into the center of the spotlight became the Personal Agent, which OpenAI named Dot.
This launch event announced 25 updates, compiled by Dongcha Beating as follows.
Models and dot
Dot
A Personal Agent powered by GPT-6 Astra. It has an independent cloud sandbox computer and a dedicated browser, connected to over 4,000 external apps; supports proactive research, able to browse apps in the background with read-only permissions to find work to do; Pro and Business Premium users receive their first dot for free, and conversations do not count toward the ChatGPT base quota.
Specialist Dots
A preview version of digital employees for enterprises. They have independent identities, credentials, and employee IDs within the organization.
GPT-6.1 Sol
Focused on agent coding, computer operation, and professional tasks. Input $2 per million tokens, output $10 (only one-fifth of Astra), cached input $0.10; DeepSWE v1.1 score matches Astra, Terminal-Bench score reaches more than twice that of the previous generation Sol.
Astra Ultrafast & Sol Ultrafast
High-throughput ultra-fast tier. Speed increased by up to 8 times, Codex throughput about 300 tokens per second. Priced at 6 times the standard Astra version (input $60, output $300).
Privacy Intelligence
All tiers support "zero data retention" with offline security review; partnered with Cisco, Databricks, and Snowflake to launch privacy inference based on confidential computing architecture.
Codex and API
Codex Cloud
Fully managed cloud development environment. Supports offline operation, processes continue running even when the laptop is closed, and teams can share pre-configured containers.
New Codex CLI
Redesigned terminal interaction, supports real-time voice commands, native worktree support, and a new /agents view for multi-agent orchestration and progress tracking.
Code Review
Integrated into the desktop client, supports GitHub PR and GitLab MR, and can automatically complete initial review and diff diagnostics via the cloud even while offline.
Codex Security Cloud
Productized the internal security project Defense Factory, integrated with the Daybreak Blue cybersecurity large model, providing continuous vulnerability scanning and automated remediation for code repositories.
Decisions API
OpenAI's version of Jev, a dedicated decision interface that locks GPT-6 Luna within a fixed Q&A scope. Purpose-built for intent classification and agent path routing, with a response time of approximately 150 milliseconds and 10x faster invocation speed than conventional calls.
Agents API Opens Computer Use
Fully opens Codex harness core capabilities (multi-agent orchestration, context compression, tool retrieval, and screen operation). The framework code is open-sourced, with a fully managed option also available.
Amazon Bedrock Managed Agents
Deep integration with AWS, enabling enterprises to run OpenAI's agent systems directly within their own AWS VPC and compute resources.
Plugin System
Plugin Sidebar and File Interaction
Developers can customize dedicated interactive panels or file viewers in the ChatGPT sidebar, embedding products directly into the conversation flow.
Directory Restructuring and Passive Discovery
Revamped plugin submission and review flow, proactively recommending relevant plugins based on context during natural user conversations.
Plugin Site Integration
Supports enterprises in directly connecting self-built sites to plugins, allowing members to access internal enterprise data based on their respective permissions.
MCP Event-Driven
Integrates the MCP event specification advocated by Anthropic, enabling automatic background wake-up of plugins to draft proposals when external collaboration tools (such as project boards) undergo status changes.
Collaborative Work
ChatGPT Knowledge Space
Replaces the original knowledge base, serving as a long-term collaboration hub where team members, Codex, and dot share the same context.
Dynamic Pages
Rich text载体 collaboratively edited by multiple people and multiple agents, supporting dynamic components, real-time to-dos, and data dashboards, with support for selecting specific text to @ agents for targeted modifications.
Collaborative Slides
A presentation engine supporting real-time multi-user collaboration, capable of generating and presenting within ChatGPT, and losslessly exporting to PowerPoint or Google Slides.
Team Tasks and Scheduled Dispatch
Create periodic task flows within the enterprise that can be triggered by email, IM, or scheduled events.
Office IM Integration
Deeply integrated into Slack and Teams, enabling enterprise members to @ summon summaries or troubleshoot bugs within channels without requiring personal accounts.
Meeting Plugin
Debuting on macOS, locally records audio and combines context to generate key decisions and follow-up to-dos, with audio analyzed offline and immediately destroyed.
Personal and Team Skills Homepage
Supports centralized external display and reuse of personally built sites, plugins, and shared skills.
Monetization and Subscriptions
Sign in with ChatGPT
Unified identity authentication across the entire network. Plus and Pro users can directly consume their subscription's Token quota within third-party products; the first batch includes 16 companies such as Devin, Notion, and Vercel.
Brand-new $500 Pro tier
Offers the highest-priority concurrency quota, with exclusive access to Astra Ultrafast.
Adjustments to the original $200 Pro quota
Reopened after a 20-day suspension, but starting October 30, the compute quota for Codex and ChatGPT work has been reduced from 20x Plus to 10x, and the weekly cap for GPT-6 Pro has been halved to 100 messages. Existing users receive a $2,500 credit that expires before the end of the year.
OpenAI software marketplace
Initially partnering with 32 software vendors including Adobe, Figma, Salesforce, ServiceNow, and Harvey, enterprises can use their pre-committed spending quota with OpenAI to offset and procure third-party SaaS products in reverse.
The only real product at this DevDay is Dot.
All the other announcements, whether Sol, Ultrafast, Codex harness, Space, Pages, or MCP events, are, when taken apart, nothing but scattered components; assembled together, they are in fact Dot's bones, flesh, and nerves.
Dot thinks with Astra, operates the system with Codex harness, connects to more than 4,000 external software products through plugins, wakes itself up via MCP events when dashboards change, and then writes completed work into Pages and sends it to Teams channels.
Over the past three years, the tacit understanding of human-computer interaction has remained stuck in that little box.
First it was "you ask a question, it answers"; later it became "you assign a task, it runs it and delivers."
Dot no longer waits for you to press Enter in front of the cursor. It has its own cloud computer and browser, and after the user closes the screen, it still keeps browsing various applications in the background with restricted read-only permissions, looking for whose invoice has not yet been issued and which bug still has no one fixing it.
Officials say it will finish the work your way before you even ask.
In the past, whether it was ChatGPT or various Copilots, what they competed for was seat licenses at the software level, charging monthly rent per head, essentially equipping employees with a handy screwdriver.
Dot and Specialist Dots go further. Specialist Dots have independent domain accounts, system credentials, and employee IDs within enterprises, integrated into Microsoft's Agent 365 governance system, following the familiar IT department processes for onboarding, auditing, and deactivation.
Enterprises are no longer buying a tool account; they are hiring a machine employee that doesn't require social security and never goes offline.
This is also where the two Silicon Valley giants part ways. Meta pushes its similarly shaped Muse to billions of ordinary people, leveraging the top spot on the free charts to boost consumer scale; OpenAI locks dot firmly behind the high walls of Pro and enterprise subscriptions, with more dots requiring additional monthly payments in the future.
One is grabbing users' time, the other is grabbing enterprises' workstations.
Acceleration
On the surface, it's machine employees, but behind the scenes, it's a computing bill filled with anxiety.
GPT-6.1 Sol is priced at only one-fifth of Astra, yet its performance closely rivals it on most benchmarks; Astra Ultrafast is listed at 6 times the standard price, delivering a blazing 300 tokens per second;紧接着, Sol Ultrafast is packaged as "buying close to Astra's brain and 8 times the speed at the original Astra price."
This is rare in previous model narratives. The industry used to only look at benchmark scores, but now latency and throughput are put on the shelf, becoming clearly priced luxuries.
Why?
Because for those having agents write code, a few seconds of lag can completely shatter the hard-earned flow state.
Speed has become the most expensive premium.
But OpenAI is a company with severely strained computing power.
The $200 Pro tier, due to computing power runs, had its network cable directly pulled by officials on September 10, suspending new purchases for up to 20 days; on the first day of reopening, it announced that included usage would be directly halved, from Plus's 20 times to 10 times. At the same time, a new $500 per month tier appeared, exclusively offering the fastest model.
This is a textbook example of compute rationing—reserving the scarcest resources for the pockets that care least about price, while deploying sufficiently cheap sub-flagship models to defend the mass market.
From Astra on September 3, to Sol on the 22nd, to 6.1 Sol on the 29th—three model releases in a single month. Each launch, at its core, redraws a cost baseline for data centers.
Small-model startups are collateral damage. The Decisions API pushes the smallest Luna down to 150 milliseconds per judgment, purpose-built for classification and routing. Some see it as a precision strike against TypeSafe's decision model Jev.
Universal Account
Space, Pages, slides, team tasks, meeting plugins, plus @ChatGPT stationed inside Slack and Teams.
Piece these together, and ChatGPT no longer looks like a chatbot at all.
It looks more like a desktop operating system taking shape.
Directly ahead of it stands a long list of names: Notion, Google Workspace, Slack, and Microsoft Office. The slides feature specifically emphasizes lossless export to PowerPoint and Google Slides.
What's more interesting are the partners sitting in the VIP seats.
Notion is one of the first 16 partners for "Sign in with ChatGPT," letting users directly offset ChatGPT subscription credits within Notion; yet on the same day, OpenAI launched Pages and Space, which go head-to-head with Notion on virtually every feature point.
Figma, Adobe, and Salesforce made it onto the OpenAI software marketplace's big list, allowing enterprises to use contract credits prepaid to OpenAI to purchase their products; but on the same day, ChatGPT is gradually turning graphic design, layout, and customer ticket handling—work that originally belonged to them—into its own native capabilities.
"Sign in with ChatGPT" appears on the surface to be a passwordless login plugin, but at its core it's building a universal account for the AI era. 1.2 billion weekly active users don't need to pull out their wallets separately in each app—credits flow from OpenAI, and downstream developers settle their bills with OpenAI.
Whoever gets the money first is the real landlord.
The Chassis
In the first half of this year, the word most often invoked by the AI venture community to bolster its courage was "harness."
Entrepreneurs and investors liked to use it to soothe a creeping sense of insecurity. No matter how powerful the model, it's just an engine; the entire outer layer—the session orchestration, context compression, tool retrieval, and fault-tolerant recovery that actually let agents get work done—that's the real moat.
At the time, everyone was certain of one thing: the big tech companies were too busy stacking compute to bother with this dirty work.
This DevDay shattered that assumption. With the Agents API officially adding computer use, OpenAI pulled out the entire harness running under Codex, ChatGPT, and dot—bones and all. Open-sourced the code, offered hosting, and even brought AWS in to build it into Bedrock.
The big guys didn't just do it—they turned it into a factory-standard component. The most general-purpose, most standardized layer of agent framework was packaged up and taken in-house by the original vendor. Along with it, the developer's ecological niche slid into a delicate position.
Codex Cloud allows you to close your laptop and run offline in the cloud; code review can be done for you overnight; Security Cloud automatically scans and patches repositories around the clock; the /agents view in the new CLI lets one person direct multiple agents working in parallel.
The act of typing is being stripped away. Developers are no longer the ones writing code—they've become the ones sitting at their desks hitting enter to approve.
If general-purpose frameworks no longer constitute a moat, where can startups still go?
Either dig deep into industry-private domains that big tech can't reach, or go into the dead corners big tech won't touch due to compliance concerns.
But OpenAI clearly isn't planning to leave many gaps. On the same day as the launch event, zero data retention and private inference were rolled out together.
The doors left open for middlemen are closing one after another.
The Balance
The day before the launch event, many people waiting to see GPT-6.1 Astra came up empty.
OpenAI voluntarily halted this flagship model that was originally set to debut in October. Saachi Jain, who oversees the safety system, said the model had not fully met the release criteria in terms of "not crossing authorized boundaries and truthfully reporting operational behavior to humans."
This is a rare emergency brake.
In the large model race, everyone has grown accustomed to jumping the gun, grabbing GPUs, and competing over launch dates. At a time when nearly the entire industry is being swept up by FOMO and sprinting headlong, pulling a boxed-and-ready flagship model off the stage requires considerable resolve.
Stepping back often demands more determination than charging forward.
But this is precisely what makes Dot on stage seem only natural.
When an agent begins to have its own independent cloud computer, can autonomously browse the web around the clock, and invoke tools, the real technical threshold is no longer about squeezing out two more percentage points on a benchmark.
What's hard is delivering trust.
In every product detail of Dot, you can see the layered trade-offs the engineers have made.
Read-only proactive research eliminates the risk of accidental operations; highly sensitive actions each retain human approval; third-party services have their credentials managed by a system proxy, and passwords are never exposed to the model in plaintext.
These rules are defined in great detail, even appearing somewhat cautious.
But this is exactly the precondition for whether enterprises dare to truly hand over invoices, contracts, and code repositories. Security is no longer an ethics manifesto posted on an official website, but a tangible commercial access permit.
When Altman spoke on stage about this line of defense, his attitude was calm. He said that if alignment is treated purely as an engineering problem, deeper and more important propositions will be missed; as for the various jokes from the outside world, he said that if people need to project their anxiety onto someone or make fun of him to ease things a bit, he doesn't mind.
After three years of sprinting, the industry is beginning to realize that what determines how far a model can go has never been just the throughput within a compute cluster, but its sense of proportion when dealing with the real world.
All permissions have been carefully divided: which ones are handed to machines, and which ones are left to humans.
Just like that plain rule written at the bottom of the system manual, Dot can draft contracts for you, troubleshoot code, and connect 4,000 apps, but when it comes to changing a password, it will still quietly stop and wait for humans to do it themselves.
This is the beginning of maturity.
Welcome to join the official BlockBeats community:
Telegram Subscription Group: https://t.me/theblockbeats
Telegram Discussion Group: https://t.me/BlockBeats_App
Official Twitter Account: https://twitter.com/BlockBeatsAsia
