OpenAI DevDay 2026: Dots, GPT-6.1 Sol and What Actually Changes for Developers
OpenAI closed out DevDay 2026 with more than twenty announcements: GPT-6.1 Sol, near-Astra intelligence at a fifth of the price; dots, agents that keep working after you close the tab; ChatGPT Space and a $500-a-month plan.
The easy headline is "OpenAI announced twenty things" and the lazy one is "OpenAI copies Meta". Neither helps you decide anything. What happened on Tuesday, September 29 is easier to read by what changes in your work: which model you pick by default, what permissions you hand an agent that never switches off, where you ship your product, and what you pay every month.
What DevDay 2026 Was and How to Read Twenty Announcements at Once
September 29, in San Francisco
OpenAI held its annual event on Tuesday, September 29 in San Francisco, with Sam Altman on stage, and its own recap counts "more than 20 major announcements" spread across ChatGPT, Codex, the models and what it calls entirely new ways of working with AI. When twenty announcements fit into an hour of keynote, it pays to split them into three layers: what already existed and only grew, what opens today, and what still has no date.
The Five Blocks That Change Something for Builders
For anyone who writes code, DevDay 2026 comes down to five things: a new mid-tier model that gets close to the frontier at a fifth of the price; agents that never switch off; a shared workspace inside ChatGPT; the platform layer for agents; and a reordering of the subscription plans with a new $500-a-month ceiling.
Read also
GPT-6.1 Sol: Near Astra at a Fifth of the Price
The Prices, Side by Side
The model is served as gpt-6.1-sol, and its standard prices are $2 per million input tokens, $0.10 per million cached input tokens and $10 per million output tokens. Astra, the frontier model, costs $10 in and $50 out per million.
The number worth reading twice: Sol 6.1's list price is the same as GPT-6 Sol's, the model that arrived a week earlier and had already come down 50% from its predecessor. What actually drops now is the cache: $0.10 per million, 95% less than standard input and half of the previous cached rate. If your agent resends the same system prompt and tool list on every turn, that is where the savings live.
What a Third Party Measures and What OpenAI Claims
OpenAI says Sol 6.1 "nearly matches GPT-6 Astra's intelligence" on agentic coding, computer use and professional work. Artificial Analysis, measuring on its own, puts it at 52 on its intelligence index against 53 for Astra and 58 for Claude Opus 5.5, at a cost per task of $0.72 versus $3.26 for Astra, roughly 22%. The "one-fifth of the price" claim survives outside checking; the quality gap to the number-one model does not close.
Where It Is Available
According to OpenAI's launch post, Sol 6.1 is not in the consumer chat yet: you reach it through the API and in ChatGPT Work and Codex, across Plus, Pro, Business, Enterprise and Edu. Its Ultrafast version is listed as coming in the coming days. It is the same pattern of recent weeks: API first, product later. If you have to pick a default model, also read how the September 24 price war played out, because this week's context starts there.
Dots: Agents That Keep Working After You Close the Tab
Its Own Cloud Computer, a Browser and More Than 4,000 Apps
A dot is an agent that runs on GPT-6 Astra, has its own cloud computer and browser, and connects to more than 4,000 apps through plugins. It lives inside ChatGPT, Slack and Microsoft Teams, learns from your feedback and works toward your goals in the background while you do something else.
The analogy that explains it best: it is the difference between an intern waiting at the door for you to hand over work and one with its own office, its own key and its own access badge. What creates productivity is the key; what demands care is how many keys you hand out.
Where to Get One and What the First One Includes
Your first dot is included at no extra cost on Pro and Business Premium in eligible markets, including the entry-level Pro tier according to the Codex team, and conversations with your dot do not count toward your ChatGPT usage limits. On Enterprise, Edu and Healthcare it is in beta and off by default: the workspace admin turns it on. No price was announced for extra dots, so do not invent one.
Permissions: the Risk Surface of an Always-On Agent
OpenAI's safety documentation describes read-only tools for proactive research, automatic review, and sensitive tasks such as changing a password that always stay with you. The company adds that dots can still make mistakes and that you should review consequential work. The practical consequence is blunt: the question of the month stops being "which model" and becomes which permissions, what network egress and what logging you leave behind an agent that never switches off. The incident where agents posted 53 user images online is the reminder of why this section matters.
ChatGPT Space, Pages and Work Moves Into the Chat
ChatGPT Space is a shared home for a team, its dots and ChatGPT; Pages is a new type of document built for humans and agents to collaborate in, and collaborative slides with export to PowerPoint or Google Slides arrive "in the coming weeks" according to the event recap. Add @ChatGPT mentions in Slack and Teams, meeting notes, and an enterprise marketplace with 32 initial partners, among them Adobe, Figma, CrowdStrike and Harvey. For an indie developer the reading is uncomfortable and clear: ChatGPT stops being only a consumer of your API and becomes a distribution channel too, the same move Meta already rehearsed on the enterprise side with its enterprise AI platform.
What Developers Get: Agents API, Decisions API and Codex
The Agents API brings Codex's multi-agent loop, tool search and tool calling, and context compaction into your application, with OpenAI hosting the infrastructure, and now includes computer use. If you maintain your own agent loop with a scheduler, a tool router and a context-trimming step, this is OpenAI offering to host all three.
The Decisions API focuses GPT-6 Luna on questions you define with finite, pre-defined answers: classify content, route a request, pick an agent's next action. OpenAI says it decides about ten times faster than Luna does through the regular API. It is in limited preview, with broader release "in the coming days" and no separate price yet. It is a direct competitor to TypeSafe's typed-decision model.
Alongside it, Codex gains cloud execution, reviews of GitHub pull requests and GitLab merge requests from the desktop app, and Codex Security Cloud, which scans whole GitHub repositories on demand or on a schedule, keeps checking new commits and prepares fixes in the cloud. Bedrock Managed Agents also shows up for teams whose agents must run entirely inside AWS.
The Pricing Shift: Pro 500 and the $200 Plan From October 30
The new plan costs $500 a month, offers the highest allowance at 25 times the ChatGPT Plus allowance, and includes Ultrafast, the paid speed tier: up to 300 tokens per second, "up to 8x faster in Codex and up to 6x in the API", at six times the standard price, which on Astra means $60 in and $300 out per million tokens.
The $200 plan reopened for sign-ups and keeps its slot, but from October 30 its included usage in ChatGPT Work and Codex drops from 20 to 10 times the Plus allowance, and the Pro-mode message limit in chat falls from 200 to 100 per week. Put another way: the same money buys half the work a month from now. If your team pays ChatGPT bills, October 30 is the date to write down.
What to Do on Monday
Before Migrating to Sol 6.1: Measure Cost per Task, Not Price per Token
Comparing models by price per token is like comparing cars by the price of a litre: what decides the bill is how much each trip consumes. With Sol 6.1 the immediate lever is caching the stable prefix of your prompt, which is where the price really falls. Before moving a whole pipeline, measure cost per task with your own cases, not with someone else's table.
If You Leave an Agent Running: Least Privilege and Logging
An always-on agent is an always-on surface. Before handing over keys, decide what it can read, what it can execute, where it can go on the network, what lands in the log and who reviews what it did while you were away.
If You Want to Ship Inside ChatGPT
Plugin extensions, the Plugin Creator and the marketplace turn ChatGPT into one more channel, with a detail that changes the business model: through "Sign in with ChatGPT", 16 launch partners let the user's own plan pay for usage inside their app. It is worth checking whether your product fits as a plugin before building another website.
What Is Still Unknown
Most of Sol 6.1's agentic coding numbers come from OpenAI; the third-party index agrees on the price gap, not on the "near Astra" label. For Ultrafast the relative price is known, not its general availability on Sol. Some announcements shipped with no price or plan at all, such as the shared workspace and the marketplace, and two live demos stalled. All of it will move in the coming weeks.
Conclusion
DevDay 2026 produced fewer "new model" headlines than it seems: the mid-tier model closed in on the frontier without costing more, agents left the tab behind, and distribution inside ChatGPT is becoming a product decision. If you build on the API, review your cost per task and your permissions today; the rest can wait until the numbers are public and verified. And if you want to see how these models behave on real cases, keep reading the blog: we test before we recommend.
