OpenAI DevDay 2026: What Actually Changed?

DevDay used to be about new APIs. This year, the main character was an agent that keeps working after you leave.

OpenAI's September 29 event mixed shipped products, betas, previews, and promises. They are easy to blur together. This guide separates them, and explains what each one means whether you build software or simply use ChatGPT.

What are OpenAI dots?

Dots (OpenAI styles the name in lowercase) are always-on AI agents inside ChatGPT. You give one a goal and it keeps working across sessions, on its own cloud computer and browser, using the apps you have connected. OpenAI says each dot runs on GPT-6 Astra.

In the keynote demo, a dot worked alongside Slack and Teams and handed a build job to Codex on a laptop. OpenAI's help pages say dots share plugin permissions with ChatGPT, ChatGPT Work, and Codex, and can also be given their own Slack account.

If that sounds like Meta's Muse, observers at the event, including developer Simon Willison, noticed the same thing. The direction is clear across the industry: AI moving from answering to acting.

How dots handle approvals, according to OpenAI

• OpenAI says dots are instructed to hand the most sensitive actions, such as changing a password or transferring money, back to you to finish yourself.
• Dots are instructed to pause and ask before actions like permanently deleting data or installing software, which may need your approval each time.
• Some actions, such as sending recurring messages, can be approved in advance. Approving one message does not grant ongoing permission.
• Custom Rules let you add limits, such as never sending emails. They cannot switch off core safety requirements or the separate Auto-review check.
• Proactive research, where a dot checks connected sources before you ask, cannot send messages, change content through plugins, or control a browser or computer.
• Dots are not available to users under 18.

OpenAI is also candid about the limits. Its help page says the prompt-injection protections 'help reduce the risk of malicious instructions causing an unwanted action, but they do not eliminate it.' It also says you cannot currently view or delete individual dot memories. Deleting the whole dot is how you clear its context.

GPT-6.1 Sol: near-Astra work at a lower price

OpenAI released GPT-6.1 Sol on September 29. Its model page describes it as delivering 'near-Astra performance at a lower cost for complex coding, computer use, and professional work,' and suggests comparing it with Astra on your own tasks. On stage, OpenAI called it 'near-Astra level intelligence at a fifth of the price,' according to Simon Willison's live blog.

From OpenAI's docs: a 1,050,000-token context window, up to 128,000 output tokens, and an April 30, 2026 knowledge cutoff. It takes text and images in and returns text. Tool calling requires the Responses API.

In ChatGPT, OpenAI is rolling it out in ChatGPT Work and Codex, starting with Pro and expanding to Plus, Business, Enterprise and Edu.

A few pricing details are worth knowing. OpenAI says prompts over 272K input tokens are billed at 2x input and cache rates and 1.5x output for the whole request. Batch and Flex are 50% cheaper than Standard, and Fast mode costs 2x.

Compared with GPT-6 Sol, released a week earlier, input and output prices are the same, and cached input is half the price: $0.10 per million instead of $0.20, according to OpenAI's changelog.

The Agents API and computer use

OpenAI released the Agents API in public beta on September 15. It lets developers build agents on a managed Codex harness, with OpenAI handling session orchestration, context compaction, and recovery. Developers can connect their own tools and MCP servers, and run agents in OpenAI-hosted sandboxes or in a sandbox from their own infrastructure or a supported provider.

On September 29, OpenAI added computer use. Its changelog says agents can complete tasks in an OpenAI-hosted browser, 'with website access approvals and sign-in handled by your application.'

That last clause matters. Your application, not the model, decides which sites an agent may visit and how it signs in. Computer use is powerful and still in beta, so build approval steps in from the start.

OpenAI also previewed a Decisions API, in limited preview, for cases where a model chooses from a predefined set of options, such as classification or routing. If you read our piece on Jev, it is a similar idea.

Other DevDay launches worth knowing

• Ultrafast for GPT-6 Astra: Ultrafast mode is not new, but DevDay added it for GPT-6 Astra in the Responses API. OpenAI said it is up to 8x faster in Codex and up to 6x faster in the API, priced at 6x standard rates, with GPT-6.1 Sol support coming soon.
• Pro 500: a new $500-a-month ChatGPT plan that OpenAI says includes Ultrafast access in ChatGPT Work and Codex. The keynote put its usage at 25x that of Plus.
• ChatGPT Space and Pages: shared workspaces and documents where teams and their dots work together, per launch coverage.
• Sign in with ChatGPT, plugin extensions, and the OpenAI Marketplace: new ways for developers to reach ChatGPT users. OpenAI said 1.2 billion people use ChatGPT every week.
• Codex in the cloud and Codex Security Cloud: Codex tasks that keep running remotely and, OpenAI says, scans of GitHub repositories, on demand or on a schedule, that prepare proposed fixes for review.

What OpenAI DevDay 2026 means for developers

• Test GPT-6.1 Sol against Astra on your real tasks. OpenAI itself frames this as a quality-versus-cost trade-off.
• Treat the Agents API and computer use as beta. Expect changes, and keep fallbacks for anything critical.
• Own the boundaries. Allowed websites, sign-in, approvals, and logs are your application's job.
• Watch long-prompt pricing. Above 272K input tokens, the whole request is billed at higher rates.

What it means for everyday ChatGPT users

If you are on Free or Go, most of DevDay does not change your ChatGPT today. Plus users are next in line for GPT-6.1 Sol in ChatGPT Work and Codex, after Pro, but dots are not available on Plus.

Dots are limited to Pro and Business Premium in eligible markets, and OpenAI says Pro users in the EEA, UK, and Switzerland are excluded for now. OpenAI also says dots usage will not count toward plan allowances for the first month after launch.

If you do get a dot, start it on read-only research, keep approvals on, and check its Activity View in the desktop app. The same principles from our AI agent security guide apply: give it the fewest keys that do the job.