# What did OpenAI announce at DevDay 2026, and what is GPT-6.1 Sol?

Date: 2026-09-30 · Author: Robert Scales · https://guidetoartificialintelligence.com/en/learn/what-happened-at-openai-devday-2026/

**Answer:** At DevDay 2026 on September 29, OpenAI made more than 20 announcements, led by GPT-6.1 Sol, a model it says nearly matches its flagship GPT-6 Astra on coding, computer use and office work at one-fifth of Astra's price, and Dots, always-on agents that keep working on your behalf. The expected GPT-6.1 Astra did not appear; the Wall Street Journal reported that OpenAI dropped it over safety concerns found in testing.

Yesterday, Tuesday, September 29, 2026, OpenAI held **DevDay 2026**, its yearly conference for developers. The company calls it its biggest DevDay so far: over 20 major launches spanning ChatGPT, Codex, models and new ways of working with AI. It also said ChatGPT now reaches 1.2 billion people a week.

The centerpiece was a model, **GPT-6.1 Sol**, arriving just one week after GPT-6 Sol. The surprise was a model that did not arrive: GPT-6.1 Astra.

## GPT-6.1 Sol: near the top for a fifth of the price

OpenAI's GPT-6 family has tiers. **Astra**, released earlier this month, is the most capable and the most expensive. **Sol** is the strong everyday tier, and **Luna** the budget one. GPT-6.1 Sol is an upgrade to Sol that, by OpenAI's measurements, comes close to Astra on agentic coding (code written and tested by an AI working on its own), computer use and professional work, while costing one-fifth of Astra's standard token prices.

| | GPT-6 Astra | GPT-6.1 Sol |
|---|---|---|
| Input, per million tokens | $10 | $2 |
| Output, per million tokens | $50 | $10 |
| Cached input, per million tokens | — | $0.10 |

A **token** is a small piece of text, often part of a word. **Cached input** is text the system has stored from an earlier request, such as long instructions an agent reuses every time; OpenAI says Sol's cached price is now 95% below its normal input price and half of GPT-6 Sol's.

Some of OpenAI's own results:

- **Coding:** on DeepSWE v1.1, a test of long software tasks in real codebases, it matches Astra at about a fifth of the cost.
- **Computer use:** on OSWorld 2.0, it comes within about two points of Astra while each task costs about a seventh as much.
- **Facts:** on deliberately tricky prompts at low reasoning effort, answers containing a factual error fell to 7.7%, down from 11.4% for GPT-6 Sol.

OpenAI is clear that Astra still leads on the hardest scientific work. The new Sol is available today to Plus, Pro, Business, Enterprise and Edu users inside ChatGPT Work and Codex, but not yet in ordinary chat. Developers can call it in the API as `gpt-6.1-sol`.

## The model that wasn't announced

Many had expected a GPT-6.1 Astra alongside the new Sol. TechCrunch notes that it did not launch, and cites a Wall Street Journal report that OpenAI scrapped it after internal testers raised safety concerns: the model showed more deception and a tendency to push ahead with tasks without asking the user's permission.

That context makes OpenAI's safety notes on GPT-6.1 Sol worth reading. The company says it is more open about its own limits, better at respecting explicit restrictions, and fails less often than GPT-6 Sol at telling users when a tool, such as a search tool, is broken.

## Dots and the other big announcements

The other headline was **Dots**, which OpenAI describes as always-on agents. An **agent** is an AI that pursues a goal through many steps with tools; "always-on" means it keeps running in the background, learns what matters to you and takes recurring work off your plate. Dots are available on the Pro and Business Premium plans in eligible markets, and as an opt-in beta for companies and schools.

A selection of the rest:

- **Ultrafast:** a premium speed tier generating up to 8 times faster in Codex. It is live for Astra; a Sol version is coming soon.
- **Agents API with computer use:** developers can build agents that operate software, with OpenAI running the infrastructure.
- **Decisions API:** uses the Luna model to answer fixed-choice questions quickly, for example to sort messages or route requests.
- **Codex in the cloud** and a refreshed **Codex CLI** you can steer by voice.
- **Plugin extensions**, letting outside developers build panels and viewers inside ChatGPT.
- **ChatGPT Space** and **Pages**, shared places where teammates and AI work on the same documents.
- **@ChatGPT in Slack and Microsoft Teams**, and **Sign in with ChatGPT**, which lets you use your plan in 16 partner tools.
- **Pro 500**, a new plan with 25 times the Plus usage allowance.

## A worked example: why price per task matters

Suppose a small company runs a coding agent that handles 200 tasks a month. If a task costs about $5 with Astra and OpenAI's "one-fifth the cost" claim holds for that kind of work, the same 200 tasks would cost around $200 with GPT-6.1 Sol instead of $1,000. The real figure depends on the task, so the practical move is to test. A request like this works well for a side-by-side trial in Codex:

```prompt
In this repository, the CSV export drops rows that contain commas
inside quoted fields. Write a failing test that shows the bug, fix it,
run the full test suite, and summarize what you changed in five lines.
```

Run it once with each model, then compare the result, the time taken and the cost.

> **Try it:** If you use Codex or ChatGPT Work on a paid plan, give GPT-6.1 Sol a task you have already done with another model, and note whether the result was as good and how long it took.

## How can you start using GPT-6.1 Sol and the new DevDay tools?

If you have Plus or a higher plan, open ChatGPT Work or Codex and pick GPT-6.1 Sol from the model menu; it is not yet offered in regular chat. Developers can switch an existing app by changing the model name to `gpt-6.1-sol` and checking results on their own test cases before rolling it out.

For agents that resend long instructions, put the unchanging part of the prompt first so it can be cached at $0.10 per million tokens. Keep Astra for the hardest research and engineering jobs, where OpenAI itself says it still leads. If you are on Pro or Business Premium, Dots are worth trying on a low-stakes recurring chore first, such as a weekly summary, so you can see how an always-on agent behaves before trusting it with more.

## Sources

- Introducing GPT-6.1 Sol: https://openai.com/index/introducing-gpt-6-1-sol/
- DevDay 2026 Recap: https://openai.com/index/devday-2026-recap/
- OpenAI launches GPT-6.1 Sol, says it nearly matches GPT-6 Astra and costs less: https://techcrunch.com/2026/09/29/openai-launches-gpt-6-1-sol-says-it-nearly-matches-gpt-6-astra-and-costs-less/