OpenAI DevDay: Always-On Agents, GPT-6.1 Sol and a $500 Plan

At its DevDay developer conference in San Francisco on September 29, OpenAI announced more than 20 new products and features. The three that matter most: always-on agents called dots, the much cheaper GPT-6.1 Sol model, and a new pricing structure for the Pro plans, including a $500-a-month tier. For users in Europe there’s a catch that most recaps skip over.

Key takeaways

• dots are agents built on GPT-6 Astra that work around the clock, have their own cloud computer and browser, and connect to more than 4,000 apps. Pro subscribers in the European Economic Area, Switzerland and the UK won’t get them for now.
• GPT-6.1 Sol costs $2 per million input tokens and $10 per million output tokens in the API, a fifth of Astra’s prices. OpenAI says it comes close to Astra on agentic coding.
• Pro 500 is a new $500-a-month plan that includes the Ultrafast speed mode. On the existing Pro 200 plan, the included usage in ChatGPT Work and Codex is cut in half starting October 30.
• OpenAI also launched ChatGPT Space as a shared workspace, a new document type called Pages, plugin extensions for third-party developers, and a Codex that keeps running fully in the cloud.

dots: agents that don’t wait for the next prompt

Until now, ChatGPT works while you’re chatting with it. A dot instead takes on an ongoing task and sticks with it, including between conversations. It gets its own cloud computer with a browser and reaches connected services through plugins. When the user isn’t actively working with it, OpenAI says the agent does research on its own in the connected apps, in read-only mode. Anything beyond that is governed by Custom Rules: allow an action, block it, or require approval first. Some steps, such as changing a password, always stay with the human.

The idea isn’t new. Just days ago Microsoft added a similar always-on agent, Autopilot, to Copilot, along with usage-based billing. OpenAI takes a different approach on price: the first dot is included in the Pro and Business Premium plans at no extra cost, and according to Sam Altman, users will be able to pay for more. Chatting with the agent doesn’t count against usage limits, but the tasks it runs in Codex or ChatGPT Work do. So if you keep a dot busy, it draws down your plan.

The key detail for Europe comes from a statement OpenAI gave to the press: for Pro subscribers, the European Economic Area, Switzerland and the UK are not eligible markets for now. Business Premium customers, by contrast, are supposed to get the agents in every region where ChatGPT is available. OpenAI gave no timeline for individual Pro users in Europe.

GPT-6.1 Sol: nearly Astra, for a fifth of the price

Just one week after GPT-6 Sol, OpenAI is shipping an improved version. In the API, GPT-6.1 Sol costs $2 per million input tokens and $10 per million output tokens. Reused context, known as cached input, costs $0.10 per million tokens. That’s half of what GPT-6 Sol charged, and it matters most for coding agents that reread the same codebase over and over in long sessions.

So far, all performance numbers come from OpenAI itself. On the DeepSWE coding benchmark, GPT-6.1 Sol reportedly matches Astra at about a fifth of the cost. A cost breakdown from Terminal-Bench Science is especially telling: at the highest reasoning setting, an average task costs $5.47, compared with $23.21 for Anthropic’s Opus 5.5 and $23.80 for Astra. Astra still posts the best score on that test, at 68.1 percent. The model is available to Plus, Pro, Business, Enterprise and Edu customers in ChatGPT Work and Codex, but not yet in the regular ChatGPT chat window.

What’s missing is just as notable. There was no bigger Astra release. According to reports picked up by The Next Web and others, OpenAI canceled the planned October launch of GPT-6.1 Astra after it failed safety tests. That fits the picture of the past few days, in which OpenAI paused work on its most capable models after a sandbox escape.

New Pro plans: more speed at the top, less usage in the middle

The new Pro 500 plan costs $500 a month, and according to The Next Web it’s the only Pro plan with Ultrafast. That mode speeds up GPT-6 Astra in Codex by up to eight times, to as much as 300 tokens per second. It’s aimed at developers who work with a coding agent all day, for whom waiting time is real working time.

The change to the existing $200 Pro plan is less welcome. Starting October 30, the included usage in ChatGPT Work and Codex drops from 20 times to 10 times the Plus allowance, and GPT-6 Pro messages drop from 200 to 100 a week. Current subscribers keep their limits through October 29 and get a one-time usage credit worth $2,500 that expires at the end of the year. Reports differ on the exact transition, though: BGR cites OpenAI product lead Thibault Sottiaux as saying current terms continue for an unspecified period. Anyone signing up for Pro 200 today should plan on the halved allowance.

Also new is Sign in with ChatGPT: Plus and Pro subscribers will be able to spend their plan usage in 16 partner tools such as Devin, Notion and Vercel, instead of paying for a second subscription there.

Space, Pages and plugins: ChatGPT becomes a workspace

ChatGPT Space replaces the old Library and becomes a shared home for a team’s files, initially for Pro, Business and Enterprise customers on the web and in the desktop app. It holds Pages, a new document type that people, ChatGPT and dots work on together and that can update itself from connected tools. With plugin extensions, developers can build entire applications into ChatGPT, for example in the sidebar or in file viewers. Early examples come from Canva, Figma and Adobe.

For programmers, moving Codex to the cloud is the most practical change: tasks keep running while your own computer sleeps, and you can start them from your phone. For businesses, OpenAI introduced Private Intelligence, an infrastructure in which automated safety checks run without OpenAI staff seeing the content.

The bigger picture: your subscription becomes a compute budget

The conference shows where ChatGPT is headed: away from the chat window and toward agents that get work done in the background. That opens up new possibilities, but it also changes the math for users. A subscription is less and less a flat rate and more and more a compute budget that agents spend in the background. Anyone who signed up for Pro 200 because of its generous Codex allowance gets half of it starting in November, while the fastest mode sits behind the $500 threshold.

For developers, GPT-6.1 Sol is the real news of the day. If OpenAI’s numbers hold up in independent testing, the price of top-tier agentic coding drops sharply. For individual users in Europe, though, the most visible part of the keynote stays locked out for now: the dots that the presentation revolved around aren’t available to Pro subscribers there yet.