OpenAI DevDay 2026: The Creator & Solo Business Companion

Jay the AI Creative · Companion to the 6:51 video brief

What changed. What it means. What to try first.

Your job is to find one useful way to buy back time. This guide starts with the releases, then gives you three practical workflows and prompts you can adapt to your business.

Published October 6, 2026. Companion to research checked October 4; the DevDay announcement, API pricing and Sora retirement date were rechecked October 6. Access and prices can change. The workflows below are proposed experiments, not performance guarantees.

1. What was released or updated

The September 29 announcement brings together the following updates. Availability varies by product, plan and rollout. See the official DevDay brief.

Update	The brief
GPT-6.1 Sol	New API model for coding and professional work.
Dots	Ongoing agents; gradual rollout to eligible accounts.
Synced Work	Continue new synced tasks across desktop, web and mobile.
Codex Cloud	Prepared cloud environments for development tasks.
Space and Pages	Organize files, editable documents and collaboration.
Sites connected data	Supported workspace-private Sites can use visitors’ permitted app connections.
Plugin Extensions	Add interfaces such as panels, sidebar entries and file viewers.
MCP Events	Connected systems can trigger supported cloud workflows.
Agents API computer use	Hosted-browser interaction added to the Agents API.
Sign in with ChatGPT	Commercial sign-in is a limited trial; plan usage has separate eligibility.
Codex Security Cloud	Repository security work in research preview.
Ultrafast	Faster generation with model-specific eligibility and pricing.

Sol produces text; image creation uses a separate model or tool. The Sol model page explains supported inputs, output and API capabilities.

Check dot eligibility before planning around ongoing work. Steps that require your local files and tools still need the relevant computer online. Cloud and local access

2. Earlier releases: keep the dates straight

September 8: GPT Image 2.5 Sunburst and Flare launched before DevDay. They are image-generation and editing models, not video models. API changelog
September 10: Agents API entered public beta. Its computer-use addition belongs to September 29. API changelog
September 24: Sora 2 models and the Videos API reached their shutdown date. They should not be presented as new available DevDay releases. Deprecation notice
Also relevant: GPT-Live supports live voice interaction with backend reasoning and tools. Consider it when a real-time conversation is part of the job. GPT-Live documentation

3. Pick the update that fits your bottleneck

If this is slowing you down…	Start with this experiment
Turning recordings into finished content	Prepare one content package from a checked transcript.
Building a useful audience resource	Create one narrow calculator, checklist or resource page.
Keeping a launch organized	Give an eligible dot one bounded responsibility.
AI usage costs	Compare models on accepted output and total review time.

My take: start with work you already repeat. A useful workflow is easier to judge when you know how long the job takes today and what a good result looks like.

4. Workflow one: one recording, a reviewed content package

Bring: a recording you can use, a checked transcript, two or three examples of your voice, and a clear audience.

Ask for: an episode summary, three title directions, five clip candidates, a newsletter draft, social drafts and thumbnail concepts. Require evidence for factual claims and verify every clip against the recording.

Copy-and-paste prompt

Use this transcript and my approved brand examples to prepare a content package for creators and solo business owners. Preserve the speaker’s meaning. Give me three title directions, five clip candidates, one newsletter draft, three social drafts and three thumbnail concepts. Use only timestamps present in the supplied material; do not invent quotes or timestamps. Link product claims to sources or flag them for verification. End with a short review checklist.

Keep it if: the final package takes less total labor than your normal process, after checking and corrections. Track usable assets, factual errors and editing minutes.

5. Workflow two: build one tool your audience can use

Pick one problem you hear repeatedly: estimating production costs, preparing a workshop, qualifying an inquiry or organizing content. Keep the first version small enough to test by hand.

Copy-and-paste prompt

Build a creator production-cost calculator. Ask for editing hours, hourly cost, asset expenses and target profit margin as a percentage of selling price. Calculate production cost as hours × hourly cost + asset expenses, and suggested price as cost ÷ (1 − margin). Reject negative inputs and margins of 100% or more. Label the result as an estimate. Make it usable on mobile and show the calculation. Prepare a preview with tests before publishing.

Check: $100 of cost at a 20% margin should suggest $125, not $120. Test missing inputs, zero values, large values and a phone-sized screen.

Keep it if: five representative tests pass and a real target user can complete the task. Building the tool does not prove demand; customer feedback does.

For connected business data, check the Sites access model. Public resources and supported private connected-data Sites have different capabilities.

6. Workflow three: give ongoing work a clear boundary

Use a dot if your account has access. Otherwise test the same checklist manually before adding automation. Start with one launch, one source folder and one deadline.

Copy-and-paste responsibility brief

Help me keep this workshop launch current for two weeks. Use only the approved brief, calendar and sources I specify. Maintain a checklist of assets, deadlines and unresolved decisions. Prepare draft changes when the source material changes. Alert me when a deadline is at risk or a decision is needed, with links to the evidence. Keep routine updates in the checklist. Show me customer-facing drafts before sending or publishing them.

Keep it if: useful alerts arrive with evidence, missed updates decline, and correcting the checklist takes less effort than maintaining it yourself.

This is a prompt template. Reading it does not create a schedule or connect any account.

7. Understand the cost example

Assume 100 jobs, each with 10,000 uncached input tokens and 2,000 billed output tokens, using standard short-context API rates.

Model	Input / million	Output / million	Modeled total
GPT-6 Luna	$0.10	$0.50	$0.20
GPT-6.1 Sol	$2	$10	$4
GPT-6 Astra	$10	$50	$20

The calculation is one million input tokens plus 200,000 output tokens. These are illustrative API charges, separate from ChatGPT subscriptions. They exclude tools, images, hosting, retries and other charges. Reasoning and repeated agent steps can increase billed usage. Official API pricing

Equal token counts do not mean equal results. Compare the cost of a deliverable you actually accept, including your review time.

8. Your one-week test

Name the job. Choose one recurring task and one finished output.
Record the baseline. Time the job using your current process.
Supply the context. Add source material, voice examples and an acceptance checklist.
Run two examples. Log preparation, generation, checking, corrections and export time.
Make the call. Keep, revise or stop based on quality and total time saved.

Use this record for each run: task · manual minutes · AI-assisted total minutes · tool cost · corrections needed · accepted output · next change.

Time bought back = manual time − all AI-assisted work, including review and corrections. A fast first draft only helps when the finished result is useful.

9. Source desk

Use these official references when you need deeper setup or current availability:

AI handles the busywork. You keep the taste.