On September 29, 2026, OpenAI introduced new ways to build apps and delegate development work at DevDay.
You can use a ChatGPT subscription in v0, add GPT-6.1 Sol to your application, and build agents that operate a browser. Decisions API remains in limited preview, so access determines which projects you can start today.
Copy link to headingWhich DevDay features can you use now?
The announcements span models, developer APIs, and subscription features. Each has its own access requirements.
Copy link to headingBuild a prototype in v0 with your ChatGPT plan
Eligible Plus and Pro subscribers can connect their ChatGPT plan to v0. Supported OpenAI responses then draw on the ChatGPT allowance. You don't need the $500 plan to use this integration.
An internal equipment request app makes a useful first project. Colleagues can try the request and approval process with sample data before you connect company systems.
In v0's credit menu, choose Use ChatGPT plan, connect your account, approve subscription sharing, and enable the plan. Select a supported OpenAI model to begin. Image generation and delegated subagents can still consume v0 credits, so the connection doesn't cover every part of a project.
Try this prompt:
Build an internal equipment request app using sample data only. Employees should be able to request an item, explain why they need it, and see the request's status. Add a manager view with an approval queue, approve and decline actions, and a filter for requests older than five days.
Include sample requests in pending, approved, and declined states. Show an empty state when no requests match a filter. Keep changes in local state for this prototype, and include a control to switch between the employee and manager views. Don't connect company systems yet.
For an internal demo, control access to the deployment before sharing it with reviewers. A useful first deliverable is a working interface that exposes missing requirements, such as who can approve a request and what happens when equipment is unavailable.
Copy link to headingAdd GPT-6.1 Sol to an application
GPT-6.1 Sol is available through AI Gateway with the model ID openai/gpt-6.1-sol. You can call it through the AI SDK, Chat Completions, or Responses API interfaces.
One project to try is a release review assistant. Give it a pull request diff, the relevant requirements, and test output. Ask it to identify changes that lack coverage and explain which requirement each proposed test would address. That produces a review someone can assess against the source material.
With the ai package installed, this server-side function uses the AI SDK's text generation API to review the material you supply. Set an AI Gateway API key in your server's AI_GATEWAY_API_KEY environment variable before calling it.
import { generateText } from 'ai';export async function reviewRelease({ diff, requirements, testOutput }) { const { text } = await generateText({ model: 'openai/gpt-6.1-sol', prompt: ` Review this proposed release using the supplied evidence. Identify changed behavior that lacks test coverage. For each finding, cite the relevant file or diff hunk, explain the risk, and propose a test linked to a requirement. Say when the evidence is insufficient. Do not claim to have run tests. Requirements: ${requirements} Pull request diff: ${diff} Test output: ${testOutput} `, }); return text;}Pass the diff, requirements, and captured test output as strings. The function returns a review; your application supplies the repository context and runs any tests. The GPT-6.1 Sol and Luna build guide explains how to extend the workflow.
Copy link to headingDelegate a maintenance task to Codex Cloud
Codex Cloud runs coding tasks in remote environments with the repositories and tools they need. Each task gets its own workspace and can continue while your computer is asleep. You can review the work from the web, mobile, or desktop app.
A contained dependency migration makes a useful first task. Specify the package and target version, identify the affected application, and ask Codex to update callers and run the relevant tests. Require a report of any failures it couldn't resolve. Keeping the change bounded gives you a clear basis for reviewing the patch.
Prepare the cloud environment before assigning the migration, including dependency installation and any required service access. When the task finishes, inspect the changed files and test results before opening or merging a pull request.
Copy link to headingBuild an agent that tests a website through its interface
The Agents API launched in public beta before DevDay. Its computer-use addition lets an agent navigate websites and interact with their interfaces in an OpenAI-hosted browser. Your application starts a session and follows its events, including requests for website access.
For a first project, build an exploratory test of a storefront preview. Ask the agent to find an item, apply a filter, add the item to a cart, and report where the interface prevented it from continuing. Use test data and stop before checkout. Treat the report as evidence to investigate alongside your automated tests, since a successful browser session doesn't establish that every path works.
There is also an Agents API integration with Vercel Sandbox for agents that run code and work with files. That integration provides a separate choice of execution environment. The documented computer-use flow uses OpenAI's hosted browser; selecting Vercel Sandbox doesn't automatically supply that browser setup.
Copy link to headingDecide whether Astra Ultrafast fits your work
ChatGPT Pro 500 costs $500 USD per month and includes GPT-6 Astra Ultrafast in Work and Codex. Eligible Enterprise and Edu plans also have access, subject to workspace controls. Ultrafast is relevant when you spend time actively steering an agent, such as narrowing down a bug through repeated questions and code changes.
Faster generation has an allowance cost. Ultrafast consumes included usage at eight times the Standard rate, or purchased credits at six times the Standard rate. A test suite or external service can still dominate the time a task takes.
Consider it for work where waiting between interactions interrupts your progress. A maintenance task running unattended in the cloud may have less need for that speed. Pro 500's subscription allowance also remains separate from API billing for an application you build.
Copy link to headingWhere does the Decisions API fit?
The Decisions API uses Luna to answer user-defined questions with finite, predefined answers. Developers provide text or image context and use the result to classify content or choose a next action.
For a support workflow, you might define the destinations as billing, technical support, and account access. The decision selects a destination; your application then applies its routing rules. Include a route for cases that need human review when designing the workflow.
Decisions API is in limited preview. If you don't have access, you can still define the categories and prepare labeled examples, but you can't treat the API as an available dependency for a production launch.
Copy link to headingWhat else changed for developers?
ChatGPT plugin extensions let developers add sidebar entries and interactive panels to ChatGPT, including viewers for supported file types. A product with a specialized editor could use an extension to keep its editing interface beside the conversation.
DevDay also introduced dots for ongoing work and Space and Pages for collaboration. Those affect how teams organize and share work. They serve a different purpose from adding a model or an agent runtime to your own application.
Copy link to headingWhat should you account for before starting?
Keep your development budget separate from your application's operating costs. Using a ChatGPT plan inside v0 covers eligible responses while you build. It doesn't turn the resulting application's OpenAI API calls into subscription usage. ChatGPT and API billing remain separate.
Availability also varies by product. GPT-6.1 Sol being available through AI Gateway doesn't establish that it is a supported model for ChatGPT plan sharing in v0. Choose from v0's supported model list, and treat preview access as a prerequisite when planning an integration.
Copy link to headingFrequently asked questions
Copy link to headingWhat is a useful first project after OpenAI DevDay 2026?
An internal app prototype in v0 is a practical starting point if you have a workflow to improve and an eligible ChatGPT Plus or Pro plan. Use sample data and ask colleagues to try the main task before connecting live systems. For an existing codebase, a bounded Codex Cloud maintenance task gives you a patch and test results to review.
Copy link to headingDo I need ChatGPT Pro 500 to build with the DevDay releases?
No. Eligible Plus and Pro plans can connect to v0, and developers can access GPT-6.1 Sol through separately billed APIs. Pro 500 is relevant when you need its higher subscription allowance or Astra Ultrafast access in Work and Codex.
Copy link to headingDid OpenAI launch the Agents API at DevDay 2026?
The Agents API entered public beta on September 10, 2026, before DevDay. At the event on September 29, 2026, OpenAI added computer use, allowing developers to build agents that interact with websites through an OpenAI-hosted browser.
Copy link to headingCan every developer use the Decisions API now?
Decisions API remains a limited preview, so developers need access before integrating it. Teams without access can prepare their decision categories and evaluation examples without making their launch depend on the API.