OpenAI DevDay 2026: Dots agents and GPT-6.1 Sol

OpenAI introduced persistent software agents, a model priced at one-fifth of Astra list rates, and an enterprise app store.

At its DevDay 2026 event on Monday, September 29, 2026, OpenAI published an official recap detailing what it called more than 20 major announcements across ChatGPT, Codex, and API tooling. Three additions stand out for everyday technical work: an always-on agent called Dots, the GPT-6.1 Sol model, and an enterprise plugin marketplace.

OpenAI is selling these products directly, so performance multipliers and user statistics in this report come from the vendor rather than independent audits.

First: what an agent actually does

An AI agent is software designed to finish multi-step jobs on your behalf rather than merely reply inside a chat box. Always-on means tasks continue running after you close the window.

You can think of it like an assistant working at a desk while you are away. OpenAI sets this up by assigning each dot a cloud computer, connecting it to GPT-6 Astra, and plugging it into external software. The analogy breaks down quickly: unlike a human assistant, software does not inherently know when it is making an error. OpenAI has not published independent accuracy benchmarks for these tasks.

Dots in ChatGPT, Slack, and Teams

Dots runs directly inside ChatGPT, Slack, and Microsoft Teams. The initial rollout targets paid Pro and Business Premium accounts in supported countries. OpenAI includes the first dot in those tiers at no extra charge.

On Enterprise, Edu, and Healthcare plans, the beta is disabled by default. A workspace administrator must turn it on manually before anyone can use it. OpenAI claims ChatGPT now reaches 1.2 billion weekly active users, though that figure is self-reported.

GPT-6.1 Sol and the one-fifth price tag

Alongside Dots, OpenAI introduced GPT-6.1 Sol as a specialized upgrade for multi-step coding, computer operation, and office tasks. The marketing pitch is near-Astra quality at one-fifth the token price of the top tier.

A token is the basic fragment of words that a language model reads or outputs. Developers access these models through an API, which charges per million tokens processed.

On OpenAI’s standard price sheet, GPT-6 Astra costs $10 for input and $50 for output per million tokens. GPT-6.1 Sol costs $2 for input and $10 for output. The discount is real on paper, but it measures against OpenAI’s own most expensive model rather than an external competitor. The model is available in the API and for paid ChatGPT accounts from Plus through Enterprise.

OpenAI also claims an Ultrafast mode produces up to 300 tokens per second, which it describes as up to 8 times faster in Codex and up to 6 times faster in the API compared to its standard generation speed. That metric tracks raw text generation speed, not the total time needed to finish a complex project. GPT-6.1 Sol Ultrafast remains listed as coming soon.

Plugins and the enterprise marketplace

OpenAI paired the model updates with third-party software connectors. The company launched plugin extensions, a central plugin directory, and Sign in with ChatGPT, listing 16 launch partners including Devin by Cognition, Notion, and Vercel.

It also introduced the OpenAI Marketplace for corporate buyers with 32 initial partners. The roster includes Figma, Adobe, Sierra, Decagon, HubSpot, Salesforce, ServiceNow, Harvey, Legora, Palo Alto Networks, CrowdStrike, and Baseten.

Other additions mentioned in the official recap include Private Intelligence, cloud Codex, terminal CLI tools with voice control, desktop code review, and managed agents on Amazon Bedrock.

Limits and verification

Our newsroom has not independently tested Dots, GPT-6.1 Sol, or the marketplace software. The 4,000 app connections and speed benchmarks are claims made by the vendor.

Subscribers with Pro or Business Premium access can test the first dot today. Anyone relying on third-party verification for accuracy or production reliability will need to wait for independent benchmark data.