OpenAI DevDay 2026 Cut the Price of the AI Work Creators Actually Run

OpenAI announced more than 20 things at DevDay on September 29 (its own recap runs long), and the press ran with the cartoon: “dots,” a bubbly always-on agent that gets its own cloud computer and can touch 4,000 apps (9to5Mac counted the pile). For a solo creator, exactly one announcement changes what you pay to get real work done, and it isn’t the agent. It’s a quiet model called GPT-6.1 Sol that does most of what the flagship does for roughly a fifth of the price.

Here’s the split that matters, and why the shiny thing is mostly for offices.

The announcement with a dollar sign attached

Back on September 11 I wrote about GPT-6 Astra as an operator, not a generator, the model that drives your other tools rather than making media itself. Astra is powerful and, at roughly $10 per million input tokens and $50 per million output, expensive enough that running it in a loop all day was a real budget line.

GPT-6.1 Sol is the answer to that bill. It lists at $2 per million input tokens and $10 per million output, with cached input at $0.10, and OpenAI says it nearly matches Astra on agentic coding, computer use, and professional work at one-fifth of Astra’s token prices (the DevDay recap breaks the numbers down). OpenAI’s own example: a science task that cost $23.80 on Astra runs $5.47 on Sol. Same shape of work, roughly a quarter of the cost.

That is the announcement a one-person shop should read twice. Every agent workflow you’ve been rationing (drafting from research, batch-rewriting product descriptions, transcribing and tagging a back catalog, wiring a tool to do the boring 80%) just got cheaper to run without a meaningful drop in quality. These are the same back-office jobs AI already handles; the change is that running them at scale stopped being a splurge. Sol is available now in the API, in ChatGPT Work, and in Codex for Plus, Pro, Business, Enterprise, and Edu users.

One caveat is worth noting, because it says something about the pace: OpenAI quietly cancelled the planned GPT-6.1 Astra flagship after it failed internal safety testing this summer (Decrypt has the detail). The cheaper worker shipped; the more powerful operator didn’t. Read that as a reminder that “latest” is not a release schedule you control.

Dots is real, and it’s priced for a team

The headline everybody screenshotted was dots: small floating cartoon agents, each running on GPT-6 Astra with its own browser and its own set of credentials, able to plug into more than 4,000 apps and keep working in the background (TechCrunch has the full description). It’s genuinely impressive, and it is genuinely not aimed at you yet.

Dots ships only to ChatGPT Pro and Business Premium subscribers in eligible markets, with Free and Plus users left out entirely, and the EU, UK, and Switzerland excluded at launch. The Pro ladder OpenAI showed alongside it runs Pro 100 at $100 a month, Pro 200 at $200, and a new Pro 500 at $500 a month with a 25x usage allowance and access to an “Ultrafast” speed tier that generates up to 300 tokens a second at six times the normal price (the recap lays out the tiers).

For context, the whole creator tech stack I’d actually recommend runs about $100 a month and covers editing, hosting, email, and design. A $500 tier to keep an always-on avatar employed is an office expense, not a creator expense. If your business genuinely needs an agent watching a queue overnight, the math might work. For most people making videos, newsletters, and designs, dots is a demo to watch, not a subscription to buy.

What shipped that a one-person shop can use today

Strip away the office features (ChatGPT Space, shared Pages, Team Tasks, the Slack and Teams integrations, all of which assume coworkers) and a couple of things are quietly useful solo. The Codex coding harness was open-sourced on GitHub, CLI and SDK included, which matters if you build your own automations. The Agents API added a managed runtime and an OpenAI-hosted browser, so the plumbing for computer-use workflows is now less of a weekend project.

Be honest about the shape of this DevDay, though: most of it points at teams and enterprises. That is the strategic tell behind OpenAI moving into the creator layer, which it signalled last week by hiring Patreon’s founders. The consumer creator tools are coming; the infrastructure and the office suite arrived first.

The operator’s read

Running IT operations at a large telecom for two decades taught me to price the boring number before the exciting one, and here the boring number is the win. Pin the version you build on (ship Sol explicitly, not “latest,” so a model swap doesn’t rewrite your costs or your output overnight), cap your API spend, and treat any single vendor’s roadmap as a dependency you don’t own. The same instinct that says “test the cheaper model on a real task this week” also says keep your audience and your income portable, so no keynote gets to decide your business for you.

So do one thing with this news: take a workflow you abandoned because Astra-class reasoning was too expensive to loop, and re-run it on Sol. That’s where DevDay actually pays you back. The avatar can wait.